Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Coming full circle on the "makes me worry" comment I left: I asked the question in work channels about my concern and SteveL did confirm that the "S3 strong consistency" feature does apply generally to CRUD operations. I believe this means, if we assume there is exactly one RegionServer which is hosting a Region at one time, that one RegionServer is capable of ensuring that the gaps which do exist in S3 are a non-issue (without the need for an HBOSS-like solution). Taking the suggested on a file-per-store which enumerates the committed files: the RegionServer can make sure that operates which concurrently want to update that file are exclusive, e.g. a bulk load, a memstore flush, a compaction commit. On my plate today is to incorporate this into a design doc specifically for storefile metadata (from the other message in this broader thread) On 5/24/21 1:39 PM, Josh Elser wrote: I got pulled into a call with some folks from S3 at the last minute late week. There was a comment made in passing about reading the latest, written version of a file. At the moment, I didn't want to digress into that because of immutable HFiles. However, if we're tracking files-per-store in a file, that makes me worry. To the nice digging both Duo and Andrew have shared here already and Nick's point about design, I definitely think stating what we expect and mapping that to the "platforms" which provide that "today" (as we know each will change) is the only way to insulate ourselves. The Hadoop FS contract tests are also a great thing we can adopt. On 5/21/21 9:53 PM, 张铎(Duo Zhang) wrote: So maybe we could introduce a .hfilelist directory, and put the hflielist files under this directory, so we do not need to list all the files under the region directory. And considering the possible implementation for typical object storages, listing the last directory on the whole path will be less expensive. Andrew Purtell 于2021年5月22日周六 上午9:35 写道: On May 21, 2021, at 6:07 PM, 张铎 wrote: Since we just make use of the general FileSystem API to do listing, is it possible to make use of ' bucket index listing'? Yes, those words mean the same thing. Andrew Purtell 于2021年5月22日周六 上午 6:34写道: On May 20, 2021, at 4:00 AM, Wellington Chevreuil < wellington.chevre...@gmail.com> wrote: IMO it should be a file per store. Per region is not suitable here as compaction is per store. Per file means we still need to list all the files. And usually, after compaction, we need to do an atomic operation to remove several old files and add a new file, or even several files for stripe compaction. It will be easy if we just write one file to commit these changes. Fine for me if it's simpler. Mentioned the per file approach because I thought it could be easier/faster to do that, rather than having to update the store file list on every flush. AFAIK, append is out of the table, so updating this file would mean read it, write original content plus new hfile to a temp file, delete original file, rename it). That sounds right to be. A minor potential optimization is the filename could have a timestamp component, so a bucket index listing at that path would pick up a list including the latest, and the latest would be used as the manifest of valid store files. The cloud object store is expected to provide an atomic listing semantic where the file is written and closed and only then is it visible, and it is visible at once to everyone. (I think this is available on most.) Old manifest file versions could be lazily deleted. Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) < palomino...@gmail.com> escreveu: IIRC S3 is the only object storage which does not guarantee read-after-write consistency in the past... This is the quick result after googling AWS [1] Amazon S3 delivers strong read-after-write consistency automatically for all applications Azure[2] Azure Storage was designed to embrace a strong consistency model that guarantees that after the service performs an insert or update operation, subsequent read operations return the latest update. Aliyun[3] A feature requires that object operations in OSS be atomic, which indicates that operations can only either succeed or fail without intermediate states. To ensure that users can access only complete data, OSS does not return corrupted or partial data. Object operations in OSS are highly consistent. For example, when a user receives an upload (PUT) success response, the uploaded object can be read immediately, and copies of the object are written to multiple devices for redundancy. Therefore, the situations where data is not obtained when you perform the read-after-write operation do not exist. The same is true for delete operations. After you delete an object, the object and its copies no longer exist. GCP[4] Cloud Storage provides strong global consistency for the following operations,
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Just go ahead Josh, I haven't started to write the design doc yet. Thank you for your help! Josh Elser 于2021年5月25日周二 上午1:45写道: > Without completely opening Pandora's box, I will say we definitely have > multiple ways we can solve the metadata management for tracking (e.g. in > meta, in some other system table, in some other system, in a per-store > file). Each of them have pro's and con's, and each of them has "favor" > as to what pain we've most recently felt as a project. > > I don't want to defer having the discussion on what the "correct" one > should be, but I do want to point out that it's only half of the problem > of storefile tracking. > > My hope is that we can make this tracking system be pluggable, such that > we can prototype a solution that works "good enough" for now and enables > the rest of the development work to keep moving forward. > > I'm happy to see so many other folks also interested in the design of > how we store this. > > Could I suggest we move this discussion around the metadata storage into > its own thread? If Duo doesn't already have a design doc started, I can > also try to put one together this week. > > Does that work for you all? > > On 5/22/21 11:02 AM, 张铎(Duo Zhang) wrote: > > I could put up a simple design doc for this. > > > > But there is still a problem, about how to do rolling upgrading. > > > > After we changed the behavior, the region server will write partial store > > files directly into the data directory. For new region servers, this is > not > > a problem, as we will read the hfilelist file to find out the valid store > > files. > > But when rolling upgrading, we can not upgrade all the regionservers at > > once, for old regionservers, they will initialize a store by listing the > > store files, so if a new regionserver crashes when compacting and its > > regions are assigned to old regionservers, the old regionservers will be > in > > trouble... > > > > Stack 于2021年5月22日周六 下午12:14写道: > > > >> HBASE-24749 design and implementation had acknowledged compromises on > >> review: e.g. adding a new 'system table' to hold store files. I'd > suggest > >> the design and implementation need a revisit before we go forward; for > >> instance, factoring for systems other than s3 as suggested above (I like > >> the Duo list). > >> > >> S > >> > >> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > >> wrote: > >> > >>> What about just storing the hfile list in a file? Since now S3 has > strong > >>> consistency, we could safely overwrite a file then I think? > >>> > >>> And since the hfile list file will be very small, renaming will not be > a > >>> big problem. > >>> > >>> We could write the hfile list to a file called 'hfile.list.tmp', and > then > >>> rename it to 'hfile.list'. > >>> > >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we > could > >>> face that, the 'hfile.list' file is not there, but there is a > >>> 'hfile.list.tmp'. > >>> > >>> So when opening a HStore, we first check if 'hfile.list' is there, if > >> not, > >>> try 'hfile.list.tmp', rename it and load it. For safety, we could write > >> an > >>> initial hfile list file with no hfiles. So if we can not load either > >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so > >> users > >>> should try to fix it with HBCK. > >>> And in HBCK, we will do a listing and generate the 'hfile.list' file. > >>> > >>> WDYT? > >>> > >>> Thanks. > >>> > >>> Wellington Chevreuil 于2021年5月19日周三 > >>> 下午10:43写道: > >>> > Thank you, Andrew and Duo, > > Talking internally with Josh Elser, initial idea was to rebase the > >>> feature > branch with master (in order to catch with latest commits), then focus > >> on > work to have a minimal functioning hbase, in other words, together > with > >>> the > already committed work from HBASE-25391, make sure flush, compactions, > splits and merges all can take advantage of the persistent store file > manager and complete with no need to rely on renames. These all map to > >>> the > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > >>> and > validate this works well for our goals, we can then focus on > snapshots, > bulkloading and tooling. > > S3 now supports strong consistency, and I heard that they are also > > implementing atomic renaming currently, so maybe that's one of the > reasons > > why the development is silent now.. > > > Interesting, I had no idea this was being implemented. I know, > >> however, a > version of this feature is already available on latest EMR releases > (at > least from 6.2.0), and AWS team has published their own blog post with > their results: > > > >>> > >> > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > But I do not think store hfile list in meta is the only solution. It >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Oh, sorry. Missed that. I think the key point here is we should not have partial storefiles in the data directory if we want to downgrade. This is possible by setting the flag to false first to prevent new partial storefiles, and then use a HBCK command to remove all the partial storefiles? And in general, I think we should have a way to clean the broken storefiles automatically when the new layout is in use, so we could also rely on it to remove the broken storefiles maybe. Thanks. Andrew Purtell 于2021年5月25日周二 上午12:24写道: > > And for downgrading, usually we do not support downgrading from a major > version upgrading, so it is not a big problem. > > You missed an earlier comment from me. > > Our team requires this to be released in a branch-2 version or we can't use > it. Therefore I am not in favor of any solution that requires a major > version increment. > > > On Sun, May 23, 2021 at 5:43 AM 张铎(Duo Zhang) > wrote: > > > I do not think it should be a table level config. It should be a cluster > > level config. We only have one FileSystem so it is useless to let > different > > tables have different ways to store hfile list. > > > > But I think the general approach is fine. We could introduce a config for > > whether to enable 'write to data directory directly' mode. When rolling > > upgrading, the flag should be false, you can change it to true after the > > whole cluster has been upgraded. > > > > And for downgrading, usually we do not support downgrading from a major > > version upgrading, so it is not a big problem. > > > > Thanks. > > > > Andrew Purtell 于2021年5月23日周日 上午12:53写道: > > > > > Put a check in the code whether hfilelist mode or original store layout > > is > > > in use and handles both cases. Then, to upgrade: > > > > > > 1. First, perform a rolling upgrade to $NEW_VERSION . > > > > > > 2. Once upgraded to $NEW_VERSION execute an alter table command that > > > enables hfilelist mode. This will cause all regions to close and reopen > > in > > > the new mode. > > > > > > Because the rolling upgrade to $NEW_VERSION is completed first a mix of > > > old and new layouts is fine, for the brief period of time when store > > > layouts are upgrading in response to the alter command, because this > > > version can handle both. > > > > > > Downgrade to an older version is not possible after the alter table > > > command, so this must be clearly documented, but of course would not > be a > > > surprise to anyone, because the alter command is for switching to the > new > > > store layout. > > > > > > > > > > On May 22, 2021, at 8:03 AM, 张铎 wrote: > > > > > > > > I could put up a simple design doc for this. > > > > > > > > But there is still a problem, about how to do rolling upgrading. > > > > > > > > After we changed the behavior, the region server will write partial > > store > > > > files directly into the data directory. For new region servers, this > is > > > not > > > > a problem, as we will read the hfilelist file to find out the valid > > store > > > > files. > > > > But when rolling upgrading, we can not upgrade all the regionservers > at > > > > once, for old regionservers, they will initialize a store by listing > > the > > > > store files, so if a new regionserver crashes when compacting and its > > > > regions are assigned to old regionservers, the old regionservers will > > be > > > in > > > > trouble... > > > > > > > > Stack 于2021年5月22日周六 下午12:14写道: > > > > > > > >> HBASE-24749 design and implementation had acknowledged compromises > on > > > >> review: e.g. adding a new 'system table' to hold store files. I'd > > > suggest > > > >> the design and implementation need a revisit before we go forward; > for > > > >> instance, factoring for systems other than s3 as suggested above (I > > like > > > >> the Duo list). > > > >> > > > >> S > > > >> > > > >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) < > palomino...@gmail.com > > > > > > >>> wrote: > > > >>> > > > >>> What about just storing the hfile list in a file? Since now S3 has > > > strong > > > >>> consistency, we could safely overwrite a file then I think? > > > >>> > > > >>> And since the hfile list file will be very small, renaming will not > > be > > > a > > > >>> big problem. > > > >>> > > > >>> We could write the hfile list to a file called 'hfile.list.tmp', > and > > > then > > > >>> rename it to 'hfile.list'. > > > >>> > > > >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we > > > could > > > >>> face that, the 'hfile.list' file is not there, but there is a > > > >>> 'hfile.list.tmp'. > > > >>> > > > >>> So when opening a HStore, we first check if 'hfile.list' is there, > if > > > >> not, > > > >>> try 'hfile.list.tmp', rename it and load it. For safety, we could > > write > > > >> an > > > >>> initial hfile list file with no hfiles. So if we can not load > either > > > >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong > so > > > >> users > > > >>> should try to fix it with
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Without completely opening Pandora's box, I will say we definitely have multiple ways we can solve the metadata management for tracking (e.g. in meta, in some other system table, in some other system, in a per-store file). Each of them have pro's and con's, and each of them has "favor" as to what pain we've most recently felt as a project. I don't want to defer having the discussion on what the "correct" one should be, but I do want to point out that it's only half of the problem of storefile tracking. My hope is that we can make this tracking system be pluggable, such that we can prototype a solution that works "good enough" for now and enables the rest of the development work to keep moving forward. I'm happy to see so many other folks also interested in the design of how we store this. Could I suggest we move this discussion around the metadata storage into its own thread? If Duo doesn't already have a design doc started, I can also try to put one together this week. Does that work for you all? On 5/22/21 11:02 AM, 张铎(Duo Zhang) wrote: I could put up a simple design doc for this. But there is still a problem, about how to do rolling upgrading. After we changed the behavior, the region server will write partial store files directly into the data directory. For new region servers, this is not a problem, as we will read the hfilelist file to find out the valid store files. But when rolling upgrading, we can not upgrade all the regionservers at once, for old regionservers, they will initialize a store by listing the store files, so if a new regionserver crashes when compacting and its regions are assigned to old regionservers, the old regionservers will be in trouble... Stack 于2021年5月22日周六 下午12:14写道: HBASE-24749 design and implementation had acknowledged compromises on review: e.g. adding a new 'system table' to hold store files. I'd suggest the design and implementation need a revisit before we go forward; for instance, factoring for systems other than s3 as suggested above (I like the Duo list). S On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) wrote: What about just storing the hfile list in a file? Since now S3 has strong consistency, we could safely overwrite a file then I think? And since the hfile list file will be very small, renaming will not be a big problem. We could write the hfile list to a file called 'hfile.list.tmp', and then rename it to 'hfile.list'. This is safe for HDFS, and for S3, since it is not atomic, maybe we could face that, the 'hfile.list' file is not there, but there is a 'hfile.list.tmp'. So when opening a HStore, we first check if 'hfile.list' is there, if not, try 'hfile.list.tmp', rename it and load it. For safety, we could write an initial hfile list file with no hfiles. So if we can not load either 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so users should try to fix it with HBCK. And in HBCK, we will do a listing and generate the 'hfile.list' file. WDYT? Thanks. Wellington Chevreuil 于2021年5月19日周三 下午10:43写道: Thank you, Andrew and Duo, Talking internally with Josh Elser, initial idea was to rebase the feature branch with master (in order to catch with latest commits), then focus on work to have a minimal functioning hbase, in other words, together with the already committed work from HBASE-25391, make sure flush, compactions, splits and merges all can take advantage of the persistent store file manager and complete with no need to rely on renames. These all map to the substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test and validate this works well for our goals, we can then focus on snapshots, bulkloading and tooling. S3 now supports strong consistency, and I heard that they are also implementing atomic renaming currently, so maybe that's one of the reasons why the development is silent now.. Interesting, I had no idea this was being implemented. I know, however, a version of this feature is already available on latest EMR releases (at least from 6.2.0), and AWS team has published their own blog post with their results: https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ But I do not think store hfile list in meta is the only solution. It will cause cyclic dependencies for hbase:meta, and then force us a have a fallback solution which makes the code a bit ugly. We should try to see if this could be done with only the FileSystem. This is indeed a relevant concern. One idea I had mentioned in the original design doc was to track committed/non-committed files through xattr (or tags), which may have its own performance issues as explained by Stephen Wu, but is something that could be attempted. Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) < palomino...@gmail.com escreveu: S3 now supports strong consistency, and I heard that they are also implementing
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
I got pulled into a call with some folks from S3 at the last minute late week. There was a comment made in passing about reading the latest, written version of a file. At the moment, I didn't want to digress into that because of immutable HFiles. However, if we're tracking files-per-store in a file, that makes me worry. To the nice digging both Duo and Andrew have shared here already and Nick's point about design, I definitely think stating what we expect and mapping that to the "platforms" which provide that "today" (as we know each will change) is the only way to insulate ourselves. The Hadoop FS contract tests are also a great thing we can adopt. On 5/21/21 9:53 PM, 张铎(Duo Zhang) wrote: So maybe we could introduce a .hfilelist directory, and put the hflielist files under this directory, so we do not need to list all the files under the region directory. And considering the possible implementation for typical object storages, listing the last directory on the whole path will be less expensive. Andrew Purtell 于2021年5月22日周六 上午9:35写道: On May 21, 2021, at 6:07 PM, 张铎 wrote: Since we just make use of the general FileSystem API to do listing, is it possible to make use of ' bucket index listing'? Yes, those words mean the same thing. Andrew Purtell 于2021年5月22日周六 上午6:34写道: On May 20, 2021, at 4:00 AM, Wellington Chevreuil < wellington.chevre...@gmail.com> wrote: IMO it should be a file per store. Per region is not suitable here as compaction is per store. Per file means we still need to list all the files. And usually, after compaction, we need to do an atomic operation to remove several old files and add a new file, or even several files for stripe compaction. It will be easy if we just write one file to commit these changes. Fine for me if it's simpler. Mentioned the per file approach because I thought it could be easier/faster to do that, rather than having to update the store file list on every flush. AFAIK, append is out of the table, so updating this file would mean read it, write original content plus new hfile to a temp file, delete original file, rename it). That sounds right to be. A minor potential optimization is the filename could have a timestamp component, so a bucket index listing at that path would pick up a list including the latest, and the latest would be used as the manifest of valid store files. The cloud object store is expected to provide an atomic listing semantic where the file is written and closed and only then is it visible, and it is visible at once to everyone. (I think this is available on most.) Old manifest file versions could be lazily deleted. Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) < palomino...@gmail.com> escreveu: IIRC S3 is the only object storage which does not guarantee read-after-write consistency in the past... This is the quick result after googling AWS [1] Amazon S3 delivers strong read-after-write consistency automatically for all applications Azure[2] Azure Storage was designed to embrace a strong consistency model that guarantees that after the service performs an insert or update operation, subsequent read operations return the latest update. Aliyun[3] A feature requires that object operations in OSS be atomic, which indicates that operations can only either succeed or fail without intermediate states. To ensure that users can access only complete data, OSS does not return corrupted or partial data. Object operations in OSS are highly consistent. For example, when a user receives an upload (PUT) success response, the uploaded object can be read immediately, and copies of the object are written to multiple devices for redundancy. Therefore, the situations where data is not obtained when you perform the read-after-write operation do not exist. The same is true for delete operations. After you delete an object, the object and its copies no longer exist. GCP[4] Cloud Storage provides strong global consistency for the following operations, including both data and metadata: Read-after-write Read-after-metadata-update Read-after-delete Bucket listing Object listing I think these vendors could cover most end users in the world? 1. https://aws.amazon.com/cn/s3/consistency/ 2. https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet 3. https://www.alibabacloud.com/help/doc-detail/31827.htm 4. https://cloud.google.com/storage/docs/consistency Nick Dimiduk 于2021年5月19日周三 下午11:40写道: On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) wrote: What about just storing the hfile list in a file? Since now S3 has strong consistency, we could safely overwrite a file then I think? My concern is about portability. S3 isn't the only blob store in town, and consistent read-what-you-wrote semantics are not a standard feature, as far as I know. If we want something that can work on 3 or 5 major public cloud
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
The important detail is first there is an upgrade to a version that can support the new store layout across the whole cluster, so there will be no rolling upgrade related issues when the new layout is enabled. The new layout can be enabled with a new site config, a shell command to set a schema feature, whatever, that part can be hashed out in the design discussion. A cluster level configuration is fine. On Sun, May 23, 2021 at 5:43 AM 张铎(Duo Zhang) wrote: > I do not think it should be a table level config. It should be a cluster > level config. We only have one FileSystem so it is useless to let different > tables have different ways to store hfile list. > > But I think the general approach is fine. We could introduce a config for > whether to enable 'write to data directory directly' mode. When rolling > upgrading, the flag should be false, you can change it to true after the > whole cluster has been upgraded. > > And for downgrading, usually we do not support downgrading from a major > version upgrading, so it is not a big problem. > > Thanks. > > Andrew Purtell 于2021年5月23日周日 上午12:53写道: > > > Put a check in the code whether hfilelist mode or original store layout > is > > in use and handles both cases. Then, to upgrade: > > > > 1. First, perform a rolling upgrade to $NEW_VERSION . > > > > 2. Once upgraded to $NEW_VERSION execute an alter table command that > > enables hfilelist mode. This will cause all regions to close and reopen > in > > the new mode. > > > > Because the rolling upgrade to $NEW_VERSION is completed first a mix of > > old and new layouts is fine, for the brief period of time when store > > layouts are upgrading in response to the alter command, because this > > version can handle both. > > > > Downgrade to an older version is not possible after the alter table > > command, so this must be clearly documented, but of course would not be a > > surprise to anyone, because the alter command is for switching to the new > > store layout. > > > > > > > On May 22, 2021, at 8:03 AM, 张铎 wrote: > > > > > > I could put up a simple design doc for this. > > > > > > But there is still a problem, about how to do rolling upgrading. > > > > > > After we changed the behavior, the region server will write partial > store > > > files directly into the data directory. For new region servers, this is > > not > > > a problem, as we will read the hfilelist file to find out the valid > store > > > files. > > > But when rolling upgrading, we can not upgrade all the regionservers at > > > once, for old regionservers, they will initialize a store by listing > the > > > store files, so if a new regionserver crashes when compacting and its > > > regions are assigned to old regionservers, the old regionservers will > be > > in > > > trouble... > > > > > > Stack 于2021年5月22日周六 下午12:14写道: > > > > > >> HBASE-24749 design and implementation had acknowledged compromises on > > >> review: e.g. adding a new 'system table' to hold store files. I'd > > suggest > > >> the design and implementation need a revisit before we go forward; for > > >> instance, factoring for systems other than s3 as suggested above (I > like > > >> the Duo list). > > >> > > >> S > > >> > > >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > > > >>> wrote: > > >>> > > >>> What about just storing the hfile list in a file? Since now S3 has > > strong > > >>> consistency, we could safely overwrite a file then I think? > > >>> > > >>> And since the hfile list file will be very small, renaming will not > be > > a > > >>> big problem. > > >>> > > >>> We could write the hfile list to a file called 'hfile.list.tmp', and > > then > > >>> rename it to 'hfile.list'. > > >>> > > >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we > > could > > >>> face that, the 'hfile.list' file is not there, but there is a > > >>> 'hfile.list.tmp'. > > >>> > > >>> So when opening a HStore, we first check if 'hfile.list' is there, if > > >> not, > > >>> try 'hfile.list.tmp', rename it and load it. For safety, we could > write > > >> an > > >>> initial hfile list file with no hfiles. So if we can not load either > > >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so > > >> users > > >>> should try to fix it with HBCK. > > >>> And in HBCK, we will do a listing and generate the 'hfile.list' file. > > >>> > > >>> WDYT? > > >>> > > >>> Thanks. > > >>> > > >>> Wellington Chevreuil 于2021年5月19日周三 > > >>> 下午10:43写道: > > >>> > > Thank you, Andrew and Duo, > > > > Talking internally with Josh Elser, initial idea was to rebase the > > >>> feature > > branch with master (in order to catch with latest commits), then > focus > > >> on > > work to have a minimal functioning hbase, in other words, together > > with > > >>> the > > already committed work from HBASE-25391, make sure flush, > compactions, > > splits and merges all can take advantage of the persistent store > file > > manager and complete
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
> I do not think it should be a table level config. It should be a cluster level config. We only have one FileSystem so it is useless to let different tables have different ways to store hfile list. The perspective that claims this "useless" is a limited perspective. In our clusters, we value features that support incrementalism. On Sun, May 23, 2021 at 5:43 AM 张铎(Duo Zhang) wrote: > I do not think it should be a table level config. It should be a cluster > level config. We only have one FileSystem so it is useless to let different > tables have different ways to store hfile list. > > But I think the general approach is fine. We could introduce a config for > whether to enable 'write to data directory directly' mode. When rolling > upgrading, the flag should be false, you can change it to true after the > whole cluster has been upgraded. > > And for downgrading, usually we do not support downgrading from a major > version upgrading, so it is not a big problem. > > Thanks. > > Andrew Purtell 于2021年5月23日周日 上午12:53写道: > > > Put a check in the code whether hfilelist mode or original store layout > is > > in use and handles both cases. Then, to upgrade: > > > > 1. First, perform a rolling upgrade to $NEW_VERSION . > > > > 2. Once upgraded to $NEW_VERSION execute an alter table command that > > enables hfilelist mode. This will cause all regions to close and reopen > in > > the new mode. > > > > Because the rolling upgrade to $NEW_VERSION is completed first a mix of > > old and new layouts is fine, for the brief period of time when store > > layouts are upgrading in response to the alter command, because this > > version can handle both. > > > > Downgrade to an older version is not possible after the alter table > > command, so this must be clearly documented, but of course would not be a > > surprise to anyone, because the alter command is for switching to the new > > store layout. > > > > > > > On May 22, 2021, at 8:03 AM, 张铎 wrote: > > > > > > I could put up a simple design doc for this. > > > > > > But there is still a problem, about how to do rolling upgrading. > > > > > > After we changed the behavior, the region server will write partial > store > > > files directly into the data directory. For new region servers, this is > > not > > > a problem, as we will read the hfilelist file to find out the valid > store > > > files. > > > But when rolling upgrading, we can not upgrade all the regionservers at > > > once, for old regionservers, they will initialize a store by listing > the > > > store files, so if a new regionserver crashes when compacting and its > > > regions are assigned to old regionservers, the old regionservers will > be > > in > > > trouble... > > > > > > Stack 于2021年5月22日周六 下午12:14写道: > > > > > >> HBASE-24749 design and implementation had acknowledged compromises on > > >> review: e.g. adding a new 'system table' to hold store files. I'd > > suggest > > >> the design and implementation need a revisit before we go forward; for > > >> instance, factoring for systems other than s3 as suggested above (I > like > > >> the Duo list). > > >> > > >> S > > >> > > >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > > > >>> wrote: > > >>> > > >>> What about just storing the hfile list in a file? Since now S3 has > > strong > > >>> consistency, we could safely overwrite a file then I think? > > >>> > > >>> And since the hfile list file will be very small, renaming will not > be > > a > > >>> big problem. > > >>> > > >>> We could write the hfile list to a file called 'hfile.list.tmp', and > > then > > >>> rename it to 'hfile.list'. > > >>> > > >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we > > could > > >>> face that, the 'hfile.list' file is not there, but there is a > > >>> 'hfile.list.tmp'. > > >>> > > >>> So when opening a HStore, we first check if 'hfile.list' is there, if > > >> not, > > >>> try 'hfile.list.tmp', rename it and load it. For safety, we could > write > > >> an > > >>> initial hfile list file with no hfiles. So if we can not load either > > >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so > > >> users > > >>> should try to fix it with HBCK. > > >>> And in HBCK, we will do a listing and generate the 'hfile.list' file. > > >>> > > >>> WDYT? > > >>> > > >>> Thanks. > > >>> > > >>> Wellington Chevreuil 于2021年5月19日周三 > > >>> 下午10:43写道: > > >>> > > Thank you, Andrew and Duo, > > > > Talking internally with Josh Elser, initial idea was to rebase the > > >>> feature > > branch with master (in order to catch with latest commits), then > focus > > >> on > > work to have a minimal functioning hbase, in other words, together > > with > > >>> the > > already committed work from HBASE-25391, make sure flush, > compactions, > > splits and merges all can take advantage of the persistent store > file > > manager and complete with no need to rely on renames. These all map > to > > >>> the > >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
> And for downgrading, usually we do not support downgrading from a major version upgrading, so it is not a big problem. You missed an earlier comment from me. Our team requires this to be released in a branch-2 version or we can't use it. Therefore I am not in favor of any solution that requires a major version increment. On Sun, May 23, 2021 at 5:43 AM 张铎(Duo Zhang) wrote: > I do not think it should be a table level config. It should be a cluster > level config. We only have one FileSystem so it is useless to let different > tables have different ways to store hfile list. > > But I think the general approach is fine. We could introduce a config for > whether to enable 'write to data directory directly' mode. When rolling > upgrading, the flag should be false, you can change it to true after the > whole cluster has been upgraded. > > And for downgrading, usually we do not support downgrading from a major > version upgrading, so it is not a big problem. > > Thanks. > > Andrew Purtell 于2021年5月23日周日 上午12:53写道: > > > Put a check in the code whether hfilelist mode or original store layout > is > > in use and handles both cases. Then, to upgrade: > > > > 1. First, perform a rolling upgrade to $NEW_VERSION . > > > > 2. Once upgraded to $NEW_VERSION execute an alter table command that > > enables hfilelist mode. This will cause all regions to close and reopen > in > > the new mode. > > > > Because the rolling upgrade to $NEW_VERSION is completed first a mix of > > old and new layouts is fine, for the brief period of time when store > > layouts are upgrading in response to the alter command, because this > > version can handle both. > > > > Downgrade to an older version is not possible after the alter table > > command, so this must be clearly documented, but of course would not be a > > surprise to anyone, because the alter command is for switching to the new > > store layout. > > > > > > > On May 22, 2021, at 8:03 AM, 张铎 wrote: > > > > > > I could put up a simple design doc for this. > > > > > > But there is still a problem, about how to do rolling upgrading. > > > > > > After we changed the behavior, the region server will write partial > store > > > files directly into the data directory. For new region servers, this is > > not > > > a problem, as we will read the hfilelist file to find out the valid > store > > > files. > > > But when rolling upgrading, we can not upgrade all the regionservers at > > > once, for old regionservers, they will initialize a store by listing > the > > > store files, so if a new regionserver crashes when compacting and its > > > regions are assigned to old regionservers, the old regionservers will > be > > in > > > trouble... > > > > > > Stack 于2021年5月22日周六 下午12:14写道: > > > > > >> HBASE-24749 design and implementation had acknowledged compromises on > > >> review: e.g. adding a new 'system table' to hold store files. I'd > > suggest > > >> the design and implementation need a revisit before we go forward; for > > >> instance, factoring for systems other than s3 as suggested above (I > like > > >> the Duo list). > > >> > > >> S > > >> > > >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > > > >>> wrote: > > >>> > > >>> What about just storing the hfile list in a file? Since now S3 has > > strong > > >>> consistency, we could safely overwrite a file then I think? > > >>> > > >>> And since the hfile list file will be very small, renaming will not > be > > a > > >>> big problem. > > >>> > > >>> We could write the hfile list to a file called 'hfile.list.tmp', and > > then > > >>> rename it to 'hfile.list'. > > >>> > > >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we > > could > > >>> face that, the 'hfile.list' file is not there, but there is a > > >>> 'hfile.list.tmp'. > > >>> > > >>> So when opening a HStore, we first check if 'hfile.list' is there, if > > >> not, > > >>> try 'hfile.list.tmp', rename it and load it. For safety, we could > write > > >> an > > >>> initial hfile list file with no hfiles. So if we can not load either > > >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so > > >> users > > >>> should try to fix it with HBCK. > > >>> And in HBCK, we will do a listing and generate the 'hfile.list' file. > > >>> > > >>> WDYT? > > >>> > > >>> Thanks. > > >>> > > >>> Wellington Chevreuil 于2021年5月19日周三 > > >>> 下午10:43写道: > > >>> > > Thank you, Andrew and Duo, > > > > Talking internally with Josh Elser, initial idea was to rebase the > > >>> feature > > branch with master (in order to catch with latest commits), then > focus > > >> on > > work to have a minimal functioning hbase, in other words, together > > with > > >>> the > > already committed work from HBASE-25391, make sure flush, > compactions, > > splits and merges all can take advantage of the persistent store > file > > manager and complete with no need to rely on renames. These all map > to > > >>> the > > substasks
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
I do not think it should be a table level config. It should be a cluster level config. We only have one FileSystem so it is useless to let different tables have different ways to store hfile list. But I think the general approach is fine. We could introduce a config for whether to enable 'write to data directory directly' mode. When rolling upgrading, the flag should be false, you can change it to true after the whole cluster has been upgraded. And for downgrading, usually we do not support downgrading from a major version upgrading, so it is not a big problem. Thanks. Andrew Purtell 于2021年5月23日周日 上午12:53写道: > Put a check in the code whether hfilelist mode or original store layout is > in use and handles both cases. Then, to upgrade: > > 1. First, perform a rolling upgrade to $NEW_VERSION . > > 2. Once upgraded to $NEW_VERSION execute an alter table command that > enables hfilelist mode. This will cause all regions to close and reopen in > the new mode. > > Because the rolling upgrade to $NEW_VERSION is completed first a mix of > old and new layouts is fine, for the brief period of time when store > layouts are upgrading in response to the alter command, because this > version can handle both. > > Downgrade to an older version is not possible after the alter table > command, so this must be clearly documented, but of course would not be a > surprise to anyone, because the alter command is for switching to the new > store layout. > > > > On May 22, 2021, at 8:03 AM, 张铎 wrote: > > > > I could put up a simple design doc for this. > > > > But there is still a problem, about how to do rolling upgrading. > > > > After we changed the behavior, the region server will write partial store > > files directly into the data directory. For new region servers, this is > not > > a problem, as we will read the hfilelist file to find out the valid store > > files. > > But when rolling upgrading, we can not upgrade all the regionservers at > > once, for old regionservers, they will initialize a store by listing the > > store files, so if a new regionserver crashes when compacting and its > > regions are assigned to old regionservers, the old regionservers will be > in > > trouble... > > > > Stack 于2021年5月22日周六 下午12:14写道: > > > >> HBASE-24749 design and implementation had acknowledged compromises on > >> review: e.g. adding a new 'system table' to hold store files. I'd > suggest > >> the design and implementation need a revisit before we go forward; for > >> instance, factoring for systems other than s3 as suggested above (I like > >> the Duo list). > >> > >> S > >> > >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > >>> wrote: > >>> > >>> What about just storing the hfile list in a file? Since now S3 has > strong > >>> consistency, we could safely overwrite a file then I think? > >>> > >>> And since the hfile list file will be very small, renaming will not be > a > >>> big problem. > >>> > >>> We could write the hfile list to a file called 'hfile.list.tmp', and > then > >>> rename it to 'hfile.list'. > >>> > >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we > could > >>> face that, the 'hfile.list' file is not there, but there is a > >>> 'hfile.list.tmp'. > >>> > >>> So when opening a HStore, we first check if 'hfile.list' is there, if > >> not, > >>> try 'hfile.list.tmp', rename it and load it. For safety, we could write > >> an > >>> initial hfile list file with no hfiles. So if we can not load either > >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so > >> users > >>> should try to fix it with HBCK. > >>> And in HBCK, we will do a listing and generate the 'hfile.list' file. > >>> > >>> WDYT? > >>> > >>> Thanks. > >>> > >>> Wellington Chevreuil 于2021年5月19日周三 > >>> 下午10:43写道: > >>> > Thank you, Andrew and Duo, > > Talking internally with Josh Elser, initial idea was to rebase the > >>> feature > branch with master (in order to catch with latest commits), then focus > >> on > work to have a minimal functioning hbase, in other words, together > with > >>> the > already committed work from HBASE-25391, make sure flush, compactions, > splits and merges all can take advantage of the persistent store file > manager and complete with no need to rely on renames. These all map to > >>> the > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > >>> and > validate this works well for our goals, we can then focus on > snapshots, > bulkloading and tooling. > > S3 now supports strong consistency, and I heard that they are also > > implementing atomic renaming currently, so maybe that's one of the > reasons > > why the development is silent now.. > > > Interesting, I had no idea this was being implemented. I know, > >> however, a > version of this feature is already available on latest EMR releases > (at > least from 6.2.0), and AWS team has published their own
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Put a check in the code whether hfilelist mode or original store layout is in use and handles both cases. Then, to upgrade: 1. First, perform a rolling upgrade to $NEW_VERSION . 2. Once upgraded to $NEW_VERSION execute an alter table command that enables hfilelist mode. This will cause all regions to close and reopen in the new mode. Because the rolling upgrade to $NEW_VERSION is completed first a mix of old and new layouts is fine, for the brief period of time when store layouts are upgrading in response to the alter command, because this version can handle both. Downgrade to an older version is not possible after the alter table command, so this must be clearly documented, but of course would not be a surprise to anyone, because the alter command is for switching to the new store layout. > On May 22, 2021, at 8:03 AM, 张铎 wrote: > > I could put up a simple design doc for this. > > But there is still a problem, about how to do rolling upgrading. > > After we changed the behavior, the region server will write partial store > files directly into the data directory. For new region servers, this is not > a problem, as we will read the hfilelist file to find out the valid store > files. > But when rolling upgrading, we can not upgrade all the regionservers at > once, for old regionservers, they will initialize a store by listing the > store files, so if a new regionserver crashes when compacting and its > regions are assigned to old regionservers, the old regionservers will be in > trouble... > > Stack 于2021年5月22日周六 下午12:14写道: > >> HBASE-24749 design and implementation had acknowledged compromises on >> review: e.g. adding a new 'system table' to hold store files. I'd suggest >> the design and implementation need a revisit before we go forward; for >> instance, factoring for systems other than s3 as suggested above (I like >> the Duo list). >> >> S >> >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) >>> wrote: >>> >>> What about just storing the hfile list in a file? Since now S3 has strong >>> consistency, we could safely overwrite a file then I think? >>> >>> And since the hfile list file will be very small, renaming will not be a >>> big problem. >>> >>> We could write the hfile list to a file called 'hfile.list.tmp', and then >>> rename it to 'hfile.list'. >>> >>> This is safe for HDFS, and for S3, since it is not atomic, maybe we could >>> face that, the 'hfile.list' file is not there, but there is a >>> 'hfile.list.tmp'. >>> >>> So when opening a HStore, we first check if 'hfile.list' is there, if >> not, >>> try 'hfile.list.tmp', rename it and load it. For safety, we could write >> an >>> initial hfile list file with no hfiles. So if we can not load either >>> 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so >> users >>> should try to fix it with HBCK. >>> And in HBCK, we will do a listing and generate the 'hfile.list' file. >>> >>> WDYT? >>> >>> Thanks. >>> >>> Wellington Chevreuil 于2021年5月19日周三 >>> 下午10:43写道: >>> Thank you, Andrew and Duo, Talking internally with Josh Elser, initial idea was to rebase the >>> feature branch with master (in order to catch with latest commits), then focus >> on work to have a minimal functioning hbase, in other words, together with >>> the already committed work from HBASE-25391, make sure flush, compactions, splits and merges all can take advantage of the persistent store file manager and complete with no need to rely on renames. These all map to >>> the substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test >>> and validate this works well for our goals, we can then focus on snapshots, bulkloading and tooling. S3 now supports strong consistency, and I heard that they are also > implementing atomic renaming currently, so maybe that's one of the reasons > why the development is silent now.. > Interesting, I had no idea this was being implemented. I know, >> however, a version of this feature is already available on latest EMR releases (at least from 6.2.0), and AWS team has published their own blog post with their results: >>> >> https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ But I do not think store hfile list in meta is the only solution. It >> will > cause cyclic dependencies for hbase:meta, and then force us a have a > fallback solution which makes the code a bit ugly. We should try to >> see if > this could be done with only the FileSystem. > This is indeed a relevant concern. One idea I had mentioned in the >>> original design doc was to track committed/non-committed files through xattr (or tags), which may have its own performance issues as explained by >> Stephen Wu, but is something that could be attempted. Em qua., 19
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
I could put up a simple design doc for this. But there is still a problem, about how to do rolling upgrading. After we changed the behavior, the region server will write partial store files directly into the data directory. For new region servers, this is not a problem, as we will read the hfilelist file to find out the valid store files. But when rolling upgrading, we can not upgrade all the regionservers at once, for old regionservers, they will initialize a store by listing the store files, so if a new regionserver crashes when compacting and its regions are assigned to old regionservers, the old regionservers will be in trouble... Stack 于2021年5月22日周六 下午12:14写道: > HBASE-24749 design and implementation had acknowledged compromises on > review: e.g. adding a new 'system table' to hold store files. I'd suggest > the design and implementation need a revisit before we go forward; for > instance, factoring for systems other than s3 as suggested above (I like > the Duo list). > > S > > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > wrote: > > > What about just storing the hfile list in a file? Since now S3 has strong > > consistency, we could safely overwrite a file then I think? > > > > And since the hfile list file will be very small, renaming will not be a > > big problem. > > > > We could write the hfile list to a file called 'hfile.list.tmp', and then > > rename it to 'hfile.list'. > > > > This is safe for HDFS, and for S3, since it is not atomic, maybe we could > > face that, the 'hfile.list' file is not there, but there is a > > 'hfile.list.tmp'. > > > > So when opening a HStore, we first check if 'hfile.list' is there, if > not, > > try 'hfile.list.tmp', rename it and load it. For safety, we could write > an > > initial hfile list file with no hfiles. So if we can not load either > > 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so > users > > should try to fix it with HBCK. > > And in HBCK, we will do a listing and generate the 'hfile.list' file. > > > > WDYT? > > > > Thanks. > > > > Wellington Chevreuil 于2021年5月19日周三 > > 下午10:43写道: > > > > > Thank you, Andrew and Duo, > > > > > > Talking internally with Josh Elser, initial idea was to rebase the > > feature > > > branch with master (in order to catch with latest commits), then focus > on > > > work to have a minimal functioning hbase, in other words, together with > > the > > > already committed work from HBASE-25391, make sure flush, compactions, > > > splits and merges all can take advantage of the persistent store file > > > manager and complete with no need to rely on renames. These all map to > > the > > > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > > and > > > validate this works well for our goals, we can then focus on snapshots, > > > bulkloading and tooling. > > > > > > S3 now supports strong consistency, and I heard that they are also > > > > implementing atomic renaming currently, so maybe that's one of the > > > reasons > > > > why the development is silent now.. > > > > > > > Interesting, I had no idea this was being implemented. I know, > however, a > > > version of this feature is already available on latest EMR releases (at > > > least from 6.2.0), and AWS team has published their own blog post with > > > their results: > > > > > > > > > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > > > > > But I do not think store hfile list in meta is the only solution. It > will > > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > > fallback solution which makes the code a bit ugly. We should try to > see > > > if > > > > this could be done with only the FileSystem. > > > > > > > This is indeed a relevant concern. One idea I had mentioned in the > > original > > > design doc was to track committed/non-committed files through xattr (or > > > tags), which may have its own performance issues as explained by > Stephen > > > Wu, but is something that could be attempted. > > > > > > Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) < > > palomino...@gmail.com > > > > > > > escreveu: > > > > > > > S3 now supports strong consistency, and I heard that they are also > > > > implementing atomic renaming currently, so maybe that's one of the > > > reasons > > > > why the development is silent now... > > > > > > > > For me, I also think deploying hbase on cloud storage is the future, > > so I > > > > would also like to participate here. > > > > > > > > But I do not think store hfile list in meta is the only solution. It > > will > > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > > fallback solution which makes the code a bit ugly. We should try to > see > > > if > > > > this could be done with only the FileSystem. > > > > > > > > Thanks. > > > > > > > > Andrew Purtell 于2021年5月19日周三 上午8:04写道: > > > > > > > > > Wellington (and et. al), > > > > > > >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
HBASE-24749 design and implementation had acknowledged compromises on review: e.g. adding a new 'system table' to hold store files. I'd suggest the design and implementation need a revisit before we go forward; for instance, factoring for systems other than s3 as suggested above (I like the Duo list). S On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) wrote: > What about just storing the hfile list in a file? Since now S3 has strong > consistency, we could safely overwrite a file then I think? > > And since the hfile list file will be very small, renaming will not be a > big problem. > > We could write the hfile list to a file called 'hfile.list.tmp', and then > rename it to 'hfile.list'. > > This is safe for HDFS, and for S3, since it is not atomic, maybe we could > face that, the 'hfile.list' file is not there, but there is a > 'hfile.list.tmp'. > > So when opening a HStore, we first check if 'hfile.list' is there, if not, > try 'hfile.list.tmp', rename it and load it. For safety, we could write an > initial hfile list file with no hfiles. So if we can not load either > 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so users > should try to fix it with HBCK. > And in HBCK, we will do a listing and generate the 'hfile.list' file. > > WDYT? > > Thanks. > > Wellington Chevreuil 于2021年5月19日周三 > 下午10:43写道: > > > Thank you, Andrew and Duo, > > > > Talking internally with Josh Elser, initial idea was to rebase the > feature > > branch with master (in order to catch with latest commits), then focus on > > work to have a minimal functioning hbase, in other words, together with > the > > already committed work from HBASE-25391, make sure flush, compactions, > > splits and merges all can take advantage of the persistent store file > > manager and complete with no need to rely on renames. These all map to > the > > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > and > > validate this works well for our goals, we can then focus on snapshots, > > bulkloading and tooling. > > > > S3 now supports strong consistency, and I heard that they are also > > > implementing atomic renaming currently, so maybe that's one of the > > reasons > > > why the development is silent now.. > > > > > Interesting, I had no idea this was being implemented. I know, however, a > > version of this feature is already available on latest EMR releases (at > > least from 6.2.0), and AWS team has published their own blog post with > > their results: > > > > > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > > > But I do not think store hfile list in meta is the only solution. It will > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > fallback solution which makes the code a bit ugly. We should try to see > > if > > > this could be done with only the FileSystem. > > > > > This is indeed a relevant concern. One idea I had mentioned in the > original > > design doc was to track committed/non-committed files through xattr (or > > tags), which may have its own performance issues as explained by Stephen > > Wu, but is something that could be attempted. > > > > Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) < > palomino...@gmail.com > > > > > escreveu: > > > > > S3 now supports strong consistency, and I heard that they are also > > > implementing atomic renaming currently, so maybe that's one of the > > reasons > > > why the development is silent now... > > > > > > For me, I also think deploying hbase on cloud storage is the future, > so I > > > would also like to participate here. > > > > > > But I do not think store hfile list in meta is the only solution. It > will > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > fallback solution which makes the code a bit ugly. We should try to see > > if > > > this could be done with only the FileSystem. > > > > > > Thanks. > > > > > > Andrew Purtell 于2021年5月19日周三 上午8:04写道: > > > > > > > Wellington (and et. al), > > > > > > > > S3 is also an important piece of our future production plans. > > > > Unfortunately, we were unable to assist much with last year's work, > on > > > > account of being sidetracked by more immediate concerns. Fortunately, > > > this > > > > renewed interest is timely in that we have an HBase 2 project where, > if > > > > this can land in a 2.5 or a 2.6, it could be an important cost to > serve > > > > optimization, and one we could and would make use of. Therefore I > would > > > > like to restate my employer's interest in this work too. It may just > be > > > > Viraj and myself in the early days. > > > > > > > > I'm not sure how best to collaborate. We could review changes from > the > > > > original authors, new changes, and/or divide up the development > tasks. > > We > > > > can certainly offer our time for testing, and can afford the costs of > > > > testing against the S3 service. > >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
So maybe we could introduce a .hfilelist directory, and put the hflielist files under this directory, so we do not need to list all the files under the region directory. And considering the possible implementation for typical object storages, listing the last directory on the whole path will be less expensive. Andrew Purtell 于2021年5月22日周六 上午9:35写道: > > > On May 21, 2021, at 6:07 PM, 张铎 wrote: > > > > Since we just make use of the general FileSystem API to do listing, is > it > > possible to make use of ' bucket index listing'? > > Yes, those words mean the same thing. > > > > > Andrew Purtell 于2021年5月22日周六 上午6:34写道: > > > >> > >> > >>> On May 20, 2021, at 4:00 AM, Wellington Chevreuil < > >> wellington.chevre...@gmail.com> wrote: > >>> > >>> > > > IMO it should be a file per store. > Per region is not suitable here as compaction is per store. > Per file means we still need to list all the files. And usually, after > compaction, we need to do an atomic operation to remove several old > >> files > and add a new file, or even several files for stripe compaction. It > >> will be > easy if we just write one file to commit these changes. > > >>> > >>> Fine for me if it's simpler. Mentioned the per file approach because I > >>> thought it could be easier/faster to do that, rather than having to > >> update > >>> the store file list on every flush. AFAIK, append is out of the table, > so > >>> updating this file would mean read it, write original content plus new > >>> hfile to a temp file, delete original file, rename it). > >>> > >> > >> That sounds right to be. > >> > >> A minor potential optimization is the filename could have a timestamp > >> component, so a bucket index listing at that path would pick up a list > >> including the latest, and the latest would be used as the manifest of > valid > >> store files. The cloud object store is expected to provide an atomic > >> listing semantic where the file is written and closed and only then is > it > >> visible, and it is visible at once to everyone. (I think this is > available > >> on most.) Old manifest file versions could be lazily deleted. > >> > >> > Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) < > >> palomino...@gmail.com> > escreveu: > > IIRC S3 is the only object storage which does not guarantee > read-after-write consistency in the past... > > This is the quick result after googling > > AWS [1] > > > Amazon S3 delivers strong read-after-write consistency automatically > >> for > > all applications > > > Azure[2] > > > Azure Storage was designed to embrace a strong consistency model that > > guarantees that after the service performs an insert or update > >> operation, > > subsequent read operations return the latest update. > > > Aliyun[3] > > > A feature requires that object operations in OSS be atomic, which > > indicates that operations can only either succeed or fail without > > intermediate states. To ensure that users can access only complete > >> data, > > OSS does not return corrupted or partial data. > > > > Object operations in OSS are highly consistent. For example, when a > >> user > > receives an upload (PUT) success response, the uploaded object can be > read > > immediately, and copies of the object are written to multiple devices > >> for > > redundancy. Therefore, the situations where data is not obtained when > >> you > > perform the read-after-write operation do not exist. The same is true > >> for > > delete operations. After you delete an object, the object and its > >> copies > no > > longer exist. > > > > GCP[4] > > > Cloud Storage provides strong global consistency for the following > > operations, including both data and metadata: > > > > Read-after-write > > Read-after-metadata-update > > Read-after-delete > > Bucket listing > > Object listing > > > > I think these vendors could cover most end users in the world? > > 1. https://aws.amazon.com/cn/s3/consistency/ > 2. > > > >> > https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet > 3. https://www.alibabacloud.com/help/doc-detail/31827.htm > 4. https://cloud.google.com/storage/docs/consistency > > Nick Dimiduk 于2021年5月19日周三 下午11:40写道: > > > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > > > wrote: > > > >> What about just storing the hfile list in a file? Since now S3 has > strong > >> consistency, we could safely overwrite a file then I think? > >> > > > > My concern is about portability. S3 isn't the only blob store in > town, > and > > consistent read-what-you-wrote semantics are not a standard feature, > as > far > > as I know. If we want something
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
> On May 21, 2021, at 6:07 PM, 张铎 wrote: > > Since we just make use of the general FileSystem API to do listing, is it > possible to make use of ' bucket index listing'? Yes, those words mean the same thing. > > Andrew Purtell 于2021年5月22日周六 上午6:34写道: > >> >> >>> On May 20, 2021, at 4:00 AM, Wellington Chevreuil < >> wellington.chevre...@gmail.com> wrote: >>> >>> IMO it should be a file per store. Per region is not suitable here as compaction is per store. Per file means we still need to list all the files. And usually, after compaction, we need to do an atomic operation to remove several old >> files and add a new file, or even several files for stripe compaction. It >> will be easy if we just write one file to commit these changes. >>> >>> Fine for me if it's simpler. Mentioned the per file approach because I >>> thought it could be easier/faster to do that, rather than having to >> update >>> the store file list on every flush. AFAIK, append is out of the table, so >>> updating this file would mean read it, write original content plus new >>> hfile to a temp file, delete original file, rename it). >>> >> >> That sounds right to be. >> >> A minor potential optimization is the filename could have a timestamp >> component, so a bucket index listing at that path would pick up a list >> including the latest, and the latest would be used as the manifest of valid >> store files. The cloud object store is expected to provide an atomic >> listing semantic where the file is written and closed and only then is it >> visible, and it is visible at once to everyone. (I think this is available >> on most.) Old manifest file versions could be lazily deleted. >> >> Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) < >> palomino...@gmail.com> escreveu: IIRC S3 is the only object storage which does not guarantee read-after-write consistency in the past... This is the quick result after googling AWS [1] > Amazon S3 delivers strong read-after-write consistency automatically >> for > all applications Azure[2] > Azure Storage was designed to embrace a strong consistency model that > guarantees that after the service performs an insert or update >> operation, > subsequent read operations return the latest update. Aliyun[3] > A feature requires that object operations in OSS be atomic, which > indicates that operations can only either succeed or fail without > intermediate states. To ensure that users can access only complete >> data, > OSS does not return corrupted or partial data. > > Object operations in OSS are highly consistent. For example, when a >> user > receives an upload (PUT) success response, the uploaded object can be read > immediately, and copies of the object are written to multiple devices >> for > redundancy. Therefore, the situations where data is not obtained when >> you > perform the read-after-write operation do not exist. The same is true >> for > delete operations. After you delete an object, the object and its >> copies no > longer exist. > GCP[4] > Cloud Storage provides strong global consistency for the following > operations, including both data and metadata: > > Read-after-write > Read-after-metadata-update > Read-after-delete > Bucket listing > Object listing > I think these vendors could cover most end users in the world? 1. https://aws.amazon.com/cn/s3/consistency/ 2. >> https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet 3. https://www.alibabacloud.com/help/doc-detail/31827.htm 4. https://cloud.google.com/storage/docs/consistency Nick Dimiduk 于2021年5月19日周三 下午11:40写道: > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > wrote: > >> What about just storing the hfile list in a file? Since now S3 has strong >> consistency, we could safely overwrite a file then I think? >> > > My concern is about portability. S3 isn't the only blob store in town, and > consistent read-what-you-wrote semantics are not a standard feature, as far > as I know. If we want something that can work on 3 or 5 major public cloud > blobstore products as well as a smattering of on-prem technologies, we > should be selective about what features we choose to rely on as > foundational to our implementation. > > Or we are explicitly saying this will only work on S3 and we'll only > support other services when they can achieve this level of >> compatibility. > > Either way, we should be clear and up-front about what semantics we demand. > Implementing some kind of a test harness that can check compatibility would > help
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Since we just make use of the general FileSystem API to do listing, is it possible to make use of ' bucket index listing'? Andrew Purtell 于2021年5月22日周六 上午6:34写道: > > > > On May 20, 2021, at 4:00 AM, Wellington Chevreuil < > wellington.chevre...@gmail.com> wrote: > > > > > >> > >> > >> IMO it should be a file per store. > >> Per region is not suitable here as compaction is per store. > >> Per file means we still need to list all the files. And usually, after > >> compaction, we need to do an atomic operation to remove several old > files > >> and add a new file, or even several files for stripe compaction. It > will be > >> easy if we just write one file to commit these changes. > >> > > > > Fine for me if it's simpler. Mentioned the per file approach because I > > thought it could be easier/faster to do that, rather than having to > update > > the store file list on every flush. AFAIK, append is out of the table, so > > updating this file would mean read it, write original content plus new > > hfile to a temp file, delete original file, rename it). > > > > That sounds right to be. > > A minor potential optimization is the filename could have a timestamp > component, so a bucket index listing at that path would pick up a list > including the latest, and the latest would be used as the manifest of valid > store files. The cloud object store is expected to provide an atomic > listing semantic where the file is written and closed and only then is it > visible, and it is visible at once to everyone. (I think this is available > on most.) Old manifest file versions could be lazily deleted. > > > >> Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) < > palomino...@gmail.com> > >> escreveu: > >> > >> IIRC S3 is the only object storage which does not guarantee > >> read-after-write consistency in the past... > >> > >> This is the quick result after googling > >> > >> AWS [1] > >> > >>> Amazon S3 delivers strong read-after-write consistency automatically > for > >>> all applications > >> > >> > >> Azure[2] > >> > >>> Azure Storage was designed to embrace a strong consistency model that > >>> guarantees that after the service performs an insert or update > operation, > >>> subsequent read operations return the latest update. > >> > >> > >> Aliyun[3] > >> > >>> A feature requires that object operations in OSS be atomic, which > >>> indicates that operations can only either succeed or fail without > >>> intermediate states. To ensure that users can access only complete > data, > >>> OSS does not return corrupted or partial data. > >>> > >>> Object operations in OSS are highly consistent. For example, when a > user > >>> receives an upload (PUT) success response, the uploaded object can be > >> read > >>> immediately, and copies of the object are written to multiple devices > for > >>> redundancy. Therefore, the situations where data is not obtained when > you > >>> perform the read-after-write operation do not exist. The same is true > for > >>> delete operations. After you delete an object, the object and its > copies > >> no > >>> longer exist. > >>> > >> > >> GCP[4] > >> > >>> Cloud Storage provides strong global consistency for the following > >>> operations, including both data and metadata: > >>> > >>> Read-after-write > >>> Read-after-metadata-update > >>> Read-after-delete > >>> Bucket listing > >>> Object listing > >>> > >> > >> I think these vendors could cover most end users in the world? > >> > >> 1. https://aws.amazon.com/cn/s3/consistency/ > >> 2. > >> > >> > https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet > >> 3. https://www.alibabacloud.com/help/doc-detail/31827.htm > >> 4. https://cloud.google.com/storage/docs/consistency > >> > >> Nick Dimiduk 于2021年5月19日周三 下午11:40写道: > >> > >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > >>> wrote: > >>> > What about just storing the hfile list in a file? Since now S3 has > >> strong > consistency, we could safely overwrite a file then I think? > > >>> > >>> My concern is about portability. S3 isn't the only blob store in town, > >> and > >>> consistent read-what-you-wrote semantics are not a standard feature, as > >> far > >>> as I know. If we want something that can work on 3 or 5 major public > >> cloud > >>> blobstore products as well as a smattering of on-prem technologies, we > >>> should be selective about what features we choose to rely on as > >>> foundational to our implementation. > >>> > >>> Or we are explicitly saying this will only work on S3 and we'll only > >>> support other services when they can achieve this level of > compatibility. > >>> > >>> Either way, we should be clear and up-front about what semantics we > >> demand. > >>> Implementing some kind of a test harness that can check compatibility > >> would > >>> help here, a similar effort to that of defining standard behaviors of > >> HDFS > >>> implementations. > >>> > >>> I love this discussion :) > >>> > >>> And since the hfile
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
> On May 20, 2021, at 4:00 AM, Wellington Chevreuil > wrote: > > >> >> >> IMO it should be a file per store. >> Per region is not suitable here as compaction is per store. >> Per file means we still need to list all the files. And usually, after >> compaction, we need to do an atomic operation to remove several old files >> and add a new file, or even several files for stripe compaction. It will be >> easy if we just write one file to commit these changes. >> > > Fine for me if it's simpler. Mentioned the per file approach because I > thought it could be easier/faster to do that, rather than having to update > the store file list on every flush. AFAIK, append is out of the table, so > updating this file would mean read it, write original content plus new > hfile to a temp file, delete original file, rename it). > That sounds right to be. A minor potential optimization is the filename could have a timestamp component, so a bucket index listing at that path would pick up a list including the latest, and the latest would be used as the manifest of valid store files. The cloud object store is expected to provide an atomic listing semantic where the file is written and closed and only then is it visible, and it is visible at once to everyone. (I think this is available on most.) Old manifest file versions could be lazily deleted. >> Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) >> escreveu: >> >> IIRC S3 is the only object storage which does not guarantee >> read-after-write consistency in the past... >> >> This is the quick result after googling >> >> AWS [1] >> >>> Amazon S3 delivers strong read-after-write consistency automatically for >>> all applications >> >> >> Azure[2] >> >>> Azure Storage was designed to embrace a strong consistency model that >>> guarantees that after the service performs an insert or update operation, >>> subsequent read operations return the latest update. >> >> >> Aliyun[3] >> >>> A feature requires that object operations in OSS be atomic, which >>> indicates that operations can only either succeed or fail without >>> intermediate states. To ensure that users can access only complete data, >>> OSS does not return corrupted or partial data. >>> >>> Object operations in OSS are highly consistent. For example, when a user >>> receives an upload (PUT) success response, the uploaded object can be >> read >>> immediately, and copies of the object are written to multiple devices for >>> redundancy. Therefore, the situations where data is not obtained when you >>> perform the read-after-write operation do not exist. The same is true for >>> delete operations. After you delete an object, the object and its copies >> no >>> longer exist. >>> >> >> GCP[4] >> >>> Cloud Storage provides strong global consistency for the following >>> operations, including both data and metadata: >>> >>> Read-after-write >>> Read-after-metadata-update >>> Read-after-delete >>> Bucket listing >>> Object listing >>> >> >> I think these vendors could cover most end users in the world? >> >> 1. https://aws.amazon.com/cn/s3/consistency/ >> 2. >> >> https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet >> 3. https://www.alibabacloud.com/help/doc-detail/31827.htm >> 4. https://cloud.google.com/storage/docs/consistency >> >> Nick Dimiduk 于2021年5月19日周三 下午11:40写道: >> >>> On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) >>> wrote: >>> What about just storing the hfile list in a file? Since now S3 has >> strong consistency, we could safely overwrite a file then I think? >>> >>> My concern is about portability. S3 isn't the only blob store in town, >> and >>> consistent read-what-you-wrote semantics are not a standard feature, as >> far >>> as I know. If we want something that can work on 3 or 5 major public >> cloud >>> blobstore products as well as a smattering of on-prem technologies, we >>> should be selective about what features we choose to rely on as >>> foundational to our implementation. >>> >>> Or we are explicitly saying this will only work on S3 and we'll only >>> support other services when they can achieve this level of compatibility. >>> >>> Either way, we should be clear and up-front about what semantics we >> demand. >>> Implementing some kind of a test harness that can check compatibility >> would >>> help here, a similar effort to that of defining standard behaviors of >> HDFS >>> implementations. >>> >>> I love this discussion :) >>> >>> And since the hfile list file will be very small, renaming will not be a big problem. >>> >>> Would this be a file per store? A file per region? Ah. Below you imply >> it's >>> per store. >>> >>> Wellington Chevreuil 于2021年5月19日周三 下午10:43写道: > Thank you, Andrew and Duo, > > Talking internally with Josh Elser, initial idea was to rebase the feature > branch with master (in order to catch with latest commits), then >> focus
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
> > IMO it should be a file per store. > Per region is not suitable here as compaction is per store. > Per file means we still need to list all the files. And usually, after > compaction, we need to do an atomic operation to remove several old files > and add a new file, or even several files for stripe compaction. It will be > easy if we just write one file to commit these changes. > Fine for me if it's simpler. Mentioned the per file approach because I thought it could be easier/faster to do that, rather than having to update the store file list on every flush. AFAIK, append is out of the table, so updating this file would mean read it, write original content plus new hfile to a temp file, delete original file, rename it). Em qui., 20 de mai. de 2021 às 02:57, 张铎(Duo Zhang) escreveu: > IIRC S3 is the only object storage which does not guarantee > read-after-write consistency in the past... > > This is the quick result after googling > > AWS [1] > > > Amazon S3 delivers strong read-after-write consistency automatically for > > all applications > > > Azure[2] > > > Azure Storage was designed to embrace a strong consistency model that > > guarantees that after the service performs an insert or update operation, > > subsequent read operations return the latest update. > > > Aliyun[3] > > > A feature requires that object operations in OSS be atomic, which > > indicates that operations can only either succeed or fail without > > intermediate states. To ensure that users can access only complete data, > > OSS does not return corrupted or partial data. > > > > Object operations in OSS are highly consistent. For example, when a user > > receives an upload (PUT) success response, the uploaded object can be > read > > immediately, and copies of the object are written to multiple devices for > > redundancy. Therefore, the situations where data is not obtained when you > > perform the read-after-write operation do not exist. The same is true for > > delete operations. After you delete an object, the object and its copies > no > > longer exist. > > > > GCP[4] > > > Cloud Storage provides strong global consistency for the following > > operations, including both data and metadata: > > > > Read-after-write > > Read-after-metadata-update > > Read-after-delete > > Bucket listing > > Object listing > > > > I think these vendors could cover most end users in the world? > > 1. https://aws.amazon.com/cn/s3/consistency/ > 2. > > https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet > 3. https://www.alibabacloud.com/help/doc-detail/31827.htm > 4. https://cloud.google.com/storage/docs/consistency > > Nick Dimiduk 于2021年5月19日周三 下午11:40写道: > > > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > > wrote: > > > > > What about just storing the hfile list in a file? Since now S3 has > strong > > > consistency, we could safely overwrite a file then I think? > > > > > > > My concern is about portability. S3 isn't the only blob store in town, > and > > consistent read-what-you-wrote semantics are not a standard feature, as > far > > as I know. If we want something that can work on 3 or 5 major public > cloud > > blobstore products as well as a smattering of on-prem technologies, we > > should be selective about what features we choose to rely on as > > foundational to our implementation. > > > > Or we are explicitly saying this will only work on S3 and we'll only > > support other services when they can achieve this level of compatibility. > > > > Either way, we should be clear and up-front about what semantics we > demand. > > Implementing some kind of a test harness that can check compatibility > would > > help here, a similar effort to that of defining standard behaviors of > HDFS > > implementations. > > > > I love this discussion :) > > > > And since the hfile list file will be very small, renaming will not be a > > > big problem. > > > > > > > Would this be a file per store? A file per region? Ah. Below you imply > it's > > per store. > > > > Wellington Chevreuil 于2021年5月19日周三 > > > 下午10:43写道: > > > > > > > Thank you, Andrew and Duo, > > > > > > > > Talking internally with Josh Elser, initial idea was to rebase the > > > feature > > > > branch with master (in order to catch with latest commits), then > focus > > on > > > > work to have a minimal functioning hbase, in other words, together > with > > > the > > > > already committed work from HBASE-25391, make sure flush, > compactions, > > > > splits and merges all can take advantage of the persistent store file > > > > manager and complete with no need to rely on renames. These all map > to > > > the > > > > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could > test > > > and > > > > validate this works well for our goals, we can then focus on > snapshots, > > > > bulkloading and tooling. > > > > > > > > S3 now supports strong consistency, and I heard that they are also > > > > > implementing atomic renaming currently, so maybe
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
IIRC S3 is the only object storage which does not guarantee read-after-write consistency in the past... This is the quick result after googling AWS [1] > Amazon S3 delivers strong read-after-write consistency automatically for > all applications Azure[2] > Azure Storage was designed to embrace a strong consistency model that > guarantees that after the service performs an insert or update operation, > subsequent read operations return the latest update. Aliyun[3] > A feature requires that object operations in OSS be atomic, which > indicates that operations can only either succeed or fail without > intermediate states. To ensure that users can access only complete data, > OSS does not return corrupted or partial data. > > Object operations in OSS are highly consistent. For example, when a user > receives an upload (PUT) success response, the uploaded object can be read > immediately, and copies of the object are written to multiple devices for > redundancy. Therefore, the situations where data is not obtained when you > perform the read-after-write operation do not exist. The same is true for > delete operations. After you delete an object, the object and its copies no > longer exist. > GCP[4] > Cloud Storage provides strong global consistency for the following > operations, including both data and metadata: > > Read-after-write > Read-after-metadata-update > Read-after-delete > Bucket listing > Object listing > I think these vendors could cover most end users in the world? 1. https://aws.amazon.com/cn/s3/consistency/ 2. https://docs.microsoft.com/en-us/azure/storage/blobs/concurrency-manage?tabs=dotnet 3. https://www.alibabacloud.com/help/doc-detail/31827.htm 4. https://cloud.google.com/storage/docs/consistency Nick Dimiduk 于2021年5月19日周三 下午11:40写道: > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > wrote: > > > What about just storing the hfile list in a file? Since now S3 has strong > > consistency, we could safely overwrite a file then I think? > > > > My concern is about portability. S3 isn't the only blob store in town, and > consistent read-what-you-wrote semantics are not a standard feature, as far > as I know. If we want something that can work on 3 or 5 major public cloud > blobstore products as well as a smattering of on-prem technologies, we > should be selective about what features we choose to rely on as > foundational to our implementation. > > Or we are explicitly saying this will only work on S3 and we'll only > support other services when they can achieve this level of compatibility. > > Either way, we should be clear and up-front about what semantics we demand. > Implementing some kind of a test harness that can check compatibility would > help here, a similar effort to that of defining standard behaviors of HDFS > implementations. > > I love this discussion :) > > And since the hfile list file will be very small, renaming will not be a > > big problem. > > > > Would this be a file per store? A file per region? Ah. Below you imply it's > per store. > > Wellington Chevreuil 于2021年5月19日周三 > > 下午10:43写道: > > > > > Thank you, Andrew and Duo, > > > > > > Talking internally with Josh Elser, initial idea was to rebase the > > feature > > > branch with master (in order to catch with latest commits), then focus > on > > > work to have a minimal functioning hbase, in other words, together with > > the > > > already committed work from HBASE-25391, make sure flush, compactions, > > > splits and merges all can take advantage of the persistent store file > > > manager and complete with no need to rely on renames. These all map to > > the > > > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > > and > > > validate this works well for our goals, we can then focus on snapshots, > > > bulkloading and tooling. > > > > > > S3 now supports strong consistency, and I heard that they are also > > > > implementing atomic renaming currently, so maybe that's one of the > > > reasons > > > > why the development is silent now.. > > > > > > > Interesting, I had no idea this was being implemented. I know, > however, a > > > version of this feature is already available on latest EMR releases (at > > > least from 6.2.0), and AWS team has published their own blog post with > > > their results: > > > > > > > > > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > > > > > But I do not think store hfile list in meta is the only solution. It > will > > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > > fallback solution which makes the code a bit ugly. We should try to > see > > > if > > > > this could be done with only the FileSystem. > > > > > > > This is indeed a relevant concern. One idea I had mentioned in the > > original > > > design doc was to track committed/non-committed files through xattr (or > > > tags), which may have its own performance issues as explained
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Oh, just saw your last comment. IMO it should be a file per store. Per region is not suitable here as compaction is per store. Per file means we still need to list all the files. And usually, after compaction, we need to do an atomic operation to remove several old files and add a new file, or even several files for stripe compaction. It will be easy if we just write one file to commit these changes. Thanks. Nick Dimiduk 于2021年5月19日周三 下午11:40写道: > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > wrote: > > > What about just storing the hfile list in a file? Since now S3 has strong > > consistency, we could safely overwrite a file then I think? > > > > My concern is about portability. S3 isn't the only blob store in town, and > consistent read-what-you-wrote semantics are not a standard feature, as far > as I know. If we want something that can work on 3 or 5 major public cloud > blobstore products as well as a smattering of on-prem technologies, we > should be selective about what features we choose to rely on as > foundational to our implementation. > > Or we are explicitly saying this will only work on S3 and we'll only > support other services when they can achieve this level of compatibility. > > Either way, we should be clear and up-front about what semantics we demand. > Implementing some kind of a test harness that can check compatibility would > help here, a similar effort to that of defining standard behaviors of HDFS > implementations. > > I love this discussion :) > > And since the hfile list file will be very small, renaming will not be a > > big problem. > > > > Would this be a file per store? A file per region? Ah. Below you imply it's > per store. > > Wellington Chevreuil 于2021年5月19日周三 > > 下午10:43写道: > > > > > Thank you, Andrew and Duo, > > > > > > Talking internally with Josh Elser, initial idea was to rebase the > > feature > > > branch with master (in order to catch with latest commits), then focus > on > > > work to have a minimal functioning hbase, in other words, together with > > the > > > already committed work from HBASE-25391, make sure flush, compactions, > > > splits and merges all can take advantage of the persistent store file > > > manager and complete with no need to rely on renames. These all map to > > the > > > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > > and > > > validate this works well for our goals, we can then focus on snapshots, > > > bulkloading and tooling. > > > > > > S3 now supports strong consistency, and I heard that they are also > > > > implementing atomic renaming currently, so maybe that's one of the > > > reasons > > > > why the development is silent now.. > > > > > > > Interesting, I had no idea this was being implemented. I know, > however, a > > > version of this feature is already available on latest EMR releases (at > > > least from 6.2.0), and AWS team has published their own blog post with > > > their results: > > > > > > > > > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > > > > > But I do not think store hfile list in meta is the only solution. It > will > > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > > fallback solution which makes the code a bit ugly. We should try to > see > > > if > > > > this could be done with only the FileSystem. > > > > > > > This is indeed a relevant concern. One idea I had mentioned in the > > original > > > design doc was to track committed/non-committed files through xattr (or > > > tags), which may have its own performance issues as explained by > Stephen > > > Wu, but is something that could be attempted. > > > > > > Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) < > > palomino...@gmail.com > > > > > > > escreveu: > > > > > > > S3 now supports strong consistency, and I heard that they are also > > > > implementing atomic renaming currently, so maybe that's one of the > > > reasons > > > > why the development is silent now... > > > > > > > > For me, I also think deploying hbase on cloud storage is the future, > > so I > > > > would also like to participate here. > > > > > > > > But I do not think store hfile list in meta is the only solution. It > > will > > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > > fallback solution which makes the code a bit ugly. We should try to > see > > > if > > > > this could be done with only the FileSystem. > > > > > > > > Thanks. > > > > > > > > Andrew Purtell 于2021年5月19日周三 上午8:04写道: > > > > > > > > > Wellington (and et. al), > > > > > > > > > > S3 is also an important piece of our future production plans. > > > > > Unfortunately, we were unable to assist much with last year's > work, > > on > > > > > account of being sidetracked by more immediate concerns. > Fortunately, > > > > this > > > > > renewed interest is timely in that we have an HBase 2 project > where, > > if > > > > > this
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
I like the idea of tracking via files in the store. We might even do a single "hfile.commit" file for each "hfile" that got committed and has to be loaded. Once the store is opening, any hfile that doesn't have a corresponding .commit file should not be loaded, then. That discards the need for rename. Obviously relies on the strong create file consistency now supported by S3, as Nick mentioned, we would need to define that as a minimum for any object store we aim to support. And there's still the Store Engine already proposed by HBASE-25395 that uses an extra table for tracking, depending on how testing goes, we could offer that as a less efficient implementation to be used with the file system that lack such semantics. Em qua., 19 de mai. de 2021 às 17:28, Andrew Purtell < andrew.purt...@gmail.com> escreveu: > Consistent read what you wrote bucket metadata operations are standard now > for S3, Google’s GCS, and anyone who uses Ceph via its radios-gw. I think > it will be table stakes for cloud object storage. Although clients will all > see the latest metadata state for an object updated in an atomic way, this > is not a guarantee regarding views over blob contents. It may be fine but > we will have to survey the real semantics of public cloud object stores. We > can pick two or three public cloud providers - I would nominate Amazon and > Alibaba’s public cloud products - as the design targets for the initial > implementation. I like the idea of borrowing from what Hadoop did to define > the FileSystem semantics contract and conformance test suite. > > I view the current state of things as a starting point not a settled > implementation. > > Hfile tracking cannot be done in meta. Meta is not a scalable place to > store state because it cannot be split. Even the minimal state we store > there now becomes unwieldy as the number of regions and tables in a cluster > grows large. In order to take this into production we require the results > of this work to be ultimately committed to branch-2 and made available in > new minor release from there. It can’t have a design dependency on > something that either doesn’t exist or cannot be released except with a > major version increment. We don’t have a path to a releasable branch-2 > implementation of a splittable meta table. I hope we can find agreement > about this design constraint. > > > > On May 19, 2021, at 8:40 AM, Nick Dimiduk wrote: > > > > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) > wrote: > > > >> What about just storing the hfile list in a file? Since now S3 has > strong > >> consistency, we could safely overwrite a file then I think? > >> > > > > My concern is about portability. S3 isn't the only blob store in town, > and > > consistent read-what-you-wrote semantics are not a standard feature, as > far > > as I know. If we want something that can work on 3 or 5 major public > cloud > > blobstore products as well as a smattering of on-prem technologies, we > > should be selective about what features we choose to rely on as > > foundational to our implementation. > > > > Or we are explicitly saying this will only work on S3 and we'll only > > support other services when they can achieve this level of compatibility. > > > > Either way, we should be clear and up-front about what semantics we > demand. > > Implementing some kind of a test harness that can check compatibility > would > > help here, a similar effort to that of defining standard behaviors of > HDFS > > implementations. > > > > I love this discussion :) > > > > And since the hfile list file will be very small, renaming will not be a > >> big problem. > >> > > > > Would this be a file per store? A file per region? Ah. Below you imply > it's > > per store. > > > > Wellington Chevreuil 于2021年5月19日周三 > >> 下午10:43写道: > >> > >>> Thank you, Andrew and Duo, > >>> > >>> Talking internally with Josh Elser, initial idea was to rebase the > >> feature > >>> branch with master (in order to catch with latest commits), then focus > on > >>> work to have a minimal functioning hbase, in other words, together with > >> the > >>> already committed work from HBASE-25391, make sure flush, compactions, > >>> splits and merges all can take advantage of the persistent store file > >>> manager and complete with no need to rely on renames. These all map to > >> the > >>> substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > >> and > >>> validate this works well for our goals, we can then focus on snapshots, > >>> bulkloading and tooling. > >>> > >>> S3 now supports strong consistency, and I heard that they are also > implementing atomic renaming currently, so maybe that's one of the > >>> reasons > why the development is silent now.. > > >>> Interesting, I had no idea this was being implemented. I know, > however, a > >>> version of this feature is already available on latest EMR releases (at > >>> least from 6.2.0), and AWS team has published their own blog post with > >>>
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Consistent read what you wrote bucket metadata operations are standard now for S3, Google’s GCS, and anyone who uses Ceph via its radios-gw. I think it will be table stakes for cloud object storage. Although clients will all see the latest metadata state for an object updated in an atomic way, this is not a guarantee regarding views over blob contents. It may be fine but we will have to survey the real semantics of public cloud object stores. We can pick two or three public cloud providers - I would nominate Amazon and Alibaba’s public cloud products - as the design targets for the initial implementation. I like the idea of borrowing from what Hadoop did to define the FileSystem semantics contract and conformance test suite. I view the current state of things as a starting point not a settled implementation. Hfile tracking cannot be done in meta. Meta is not a scalable place to store state because it cannot be split. Even the minimal state we store there now becomes unwieldy as the number of regions and tables in a cluster grows large. In order to take this into production we require the results of this work to be ultimately committed to branch-2 and made available in new minor release from there. It can’t have a design dependency on something that either doesn’t exist or cannot be released except with a major version increment. We don’t have a path to a releasable branch-2 implementation of a splittable meta table. I hope we can find agreement about this design constraint. > On May 19, 2021, at 8:40 AM, Nick Dimiduk wrote: > > On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) wrote: > >> What about just storing the hfile list in a file? Since now S3 has strong >> consistency, we could safely overwrite a file then I think? >> > > My concern is about portability. S3 isn't the only blob store in town, and > consistent read-what-you-wrote semantics are not a standard feature, as far > as I know. If we want something that can work on 3 or 5 major public cloud > blobstore products as well as a smattering of on-prem technologies, we > should be selective about what features we choose to rely on as > foundational to our implementation. > > Or we are explicitly saying this will only work on S3 and we'll only > support other services when they can achieve this level of compatibility. > > Either way, we should be clear and up-front about what semantics we demand. > Implementing some kind of a test harness that can check compatibility would > help here, a similar effort to that of defining standard behaviors of HDFS > implementations. > > I love this discussion :) > > And since the hfile list file will be very small, renaming will not be a >> big problem. >> > > Would this be a file per store? A file per region? Ah. Below you imply it's > per store. > > Wellington Chevreuil 于2021年5月19日周三 >> 下午10:43写道: >> >>> Thank you, Andrew and Duo, >>> >>> Talking internally with Josh Elser, initial idea was to rebase the >> feature >>> branch with master (in order to catch with latest commits), then focus on >>> work to have a minimal functioning hbase, in other words, together with >> the >>> already committed work from HBASE-25391, make sure flush, compactions, >>> splits and merges all can take advantage of the persistent store file >>> manager and complete with no need to rely on renames. These all map to >> the >>> substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test >> and >>> validate this works well for our goals, we can then focus on snapshots, >>> bulkloading and tooling. >>> >>> S3 now supports strong consistency, and I heard that they are also implementing atomic renaming currently, so maybe that's one of the >>> reasons why the development is silent now.. >>> Interesting, I had no idea this was being implemented. I know, however, a >>> version of this feature is already available on latest EMR releases (at >>> least from 6.2.0), and AWS team has published their own blog post with >>> their results: >>> >>> >> https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ >>> >>> But I do not think store hfile list in meta is the only solution. It will cause cyclic dependencies for hbase:meta, and then force us a have a fallback solution which makes the code a bit ugly. We should try to see >>> if this could be done with only the FileSystem. >>> This is indeed a relevant concern. One idea I had mentioned in the >> original >>> design doc was to track committed/non-committed files through xattr (or >>> tags), which may have its own performance issues as explained by Stephen >>> Wu, but is something that could be attempted. >>> >>> Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) < >> palomino...@gmail.com >>> escreveu: >>> S3 now supports strong consistency, and I heard that they are also implementing atomic renaming currently, so
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
On Wed, May 19, 2021 at 8:19 AM 张铎(Duo Zhang) wrote: > What about just storing the hfile list in a file? Since now S3 has strong > consistency, we could safely overwrite a file then I think? > My concern is about portability. S3 isn't the only blob store in town, and consistent read-what-you-wrote semantics are not a standard feature, as far as I know. If we want something that can work on 3 or 5 major public cloud blobstore products as well as a smattering of on-prem technologies, we should be selective about what features we choose to rely on as foundational to our implementation. Or we are explicitly saying this will only work on S3 and we'll only support other services when they can achieve this level of compatibility. Either way, we should be clear and up-front about what semantics we demand. Implementing some kind of a test harness that can check compatibility would help here, a similar effort to that of defining standard behaviors of HDFS implementations. I love this discussion :) And since the hfile list file will be very small, renaming will not be a > big problem. > Would this be a file per store? A file per region? Ah. Below you imply it's per store. Wellington Chevreuil 于2021年5月19日周三 > 下午10:43写道: > > > Thank you, Andrew and Duo, > > > > Talking internally with Josh Elser, initial idea was to rebase the > feature > > branch with master (in order to catch with latest commits), then focus on > > work to have a minimal functioning hbase, in other words, together with > the > > already committed work from HBASE-25391, make sure flush, compactions, > > splits and merges all can take advantage of the persistent store file > > manager and complete with no need to rely on renames. These all map to > the > > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test > and > > validate this works well for our goals, we can then focus on snapshots, > > bulkloading and tooling. > > > > S3 now supports strong consistency, and I heard that they are also > > > implementing atomic renaming currently, so maybe that's one of the > > reasons > > > why the development is silent now.. > > > > > Interesting, I had no idea this was being implemented. I know, however, a > > version of this feature is already available on latest EMR releases (at > > least from 6.2.0), and AWS team has published their own blog post with > > their results: > > > > > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > > > But I do not think store hfile list in meta is the only solution. It will > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > fallback solution which makes the code a bit ugly. We should try to see > > if > > > this could be done with only the FileSystem. > > > > > This is indeed a relevant concern. One idea I had mentioned in the > original > > design doc was to track committed/non-committed files through xattr (or > > tags), which may have its own performance issues as explained by Stephen > > Wu, but is something that could be attempted. > > > > Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) < > palomino...@gmail.com > > > > > escreveu: > > > > > S3 now supports strong consistency, and I heard that they are also > > > implementing atomic renaming currently, so maybe that's one of the > > reasons > > > why the development is silent now... > > > > > > For me, I also think deploying hbase on cloud storage is the future, > so I > > > would also like to participate here. > > > > > > But I do not think store hfile list in meta is the only solution. It > will > > > cause cyclic dependencies for hbase:meta, and then force us a have a > > > fallback solution which makes the code a bit ugly. We should try to see > > if > > > this could be done with only the FileSystem. > > > > > > Thanks. > > > > > > Andrew Purtell 于2021年5月19日周三 上午8:04写道: > > > > > > > Wellington (and et. al), > > > > > > > > S3 is also an important piece of our future production plans. > > > > Unfortunately, we were unable to assist much with last year's work, > on > > > > account of being sidetracked by more immediate concerns. Fortunately, > > > this > > > > renewed interest is timely in that we have an HBase 2 project where, > if > > > > this can land in a 2.5 or a 2.6, it could be an important cost to > serve > > > > optimization, and one we could and would make use of. Therefore I > would > > > > like to restate my employer's interest in this work too. It may just > be > > > > Viraj and myself in the early days. > > > > > > > > I'm not sure how best to collaborate. We could review changes from > the > > > > original authors, new changes, and/or divide up the development > tasks. > > We > > > > can certainly offer our time for testing, and can afford the costs of > > > > testing against the S3 service. > > > > > > > > > > > > On Tue, May 18, 2021 at 12:16 PM Wellington Chevreuil < > > > >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
What about just storing the hfile list in a file? Since now S3 has strong consistency, we could safely overwrite a file then I think? And since the hfile list file will be very small, renaming will not be a big problem. We could write the hfile list to a file called 'hfile.list.tmp', and then rename it to 'hfile.list'. This is safe for HDFS, and for S3, since it is not atomic, maybe we could face that, the 'hfile.list' file is not there, but there is a 'hfile.list.tmp'. So when opening a HStore, we first check if 'hfile.list' is there, if not, try 'hfile.list.tmp', rename it and load it. For safety, we could write an initial hfile list file with no hfiles. So if we can not load either 'hfile.list' or 'hfile.list.tmp', then we know something is wrong so users should try to fix it with HBCK. And in HBCK, we will do a listing and generate the 'hfile.list' file. WDYT? Thanks. Wellington Chevreuil 于2021年5月19日周三 下午10:43写道: > Thank you, Andrew and Duo, > > Talking internally with Josh Elser, initial idea was to rebase the feature > branch with master (in order to catch with latest commits), then focus on > work to have a minimal functioning hbase, in other words, together with the > already committed work from HBASE-25391, make sure flush, compactions, > splits and merges all can take advantage of the persistent store file > manager and complete with no need to rely on renames. These all map to the > substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test and > validate this works well for our goals, we can then focus on snapshots, > bulkloading and tooling. > > S3 now supports strong consistency, and I heard that they are also > > implementing atomic renaming currently, so maybe that's one of the > reasons > > why the development is silent now.. > > > Interesting, I had no idea this was being implemented. I know, however, a > version of this feature is already available on latest EMR releases (at > least from 6.2.0), and AWS team has published their own blog post with > their results: > > https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ > > But I do not think store hfile list in meta is the only solution. It will > > cause cyclic dependencies for hbase:meta, and then force us a have a > > fallback solution which makes the code a bit ugly. We should try to see > if > > this could be done with only the FileSystem. > > > This is indeed a relevant concern. One idea I had mentioned in the original > design doc was to track committed/non-committed files through xattr (or > tags), which may have its own performance issues as explained by Stephen > Wu, but is something that could be attempted. > > Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) > > escreveu: > > > S3 now supports strong consistency, and I heard that they are also > > implementing atomic renaming currently, so maybe that's one of the > reasons > > why the development is silent now... > > > > For me, I also think deploying hbase on cloud storage is the future, so I > > would also like to participate here. > > > > But I do not think store hfile list in meta is the only solution. It will > > cause cyclic dependencies for hbase:meta, and then force us a have a > > fallback solution which makes the code a bit ugly. We should try to see > if > > this could be done with only the FileSystem. > > > > Thanks. > > > > Andrew Purtell 于2021年5月19日周三 上午8:04写道: > > > > > Wellington (and et. al), > > > > > > S3 is also an important piece of our future production plans. > > > Unfortunately, we were unable to assist much with last year's work, on > > > account of being sidetracked by more immediate concerns. Fortunately, > > this > > > renewed interest is timely in that we have an HBase 2 project where, if > > > this can land in a 2.5 or a 2.6, it could be an important cost to serve > > > optimization, and one we could and would make use of. Therefore I would > > > like to restate my employer's interest in this work too. It may just be > > > Viraj and myself in the early days. > > > > > > I'm not sure how best to collaborate. We could review changes from the > > > original authors, new changes, and/or divide up the development tasks. > We > > > can certainly offer our time for testing, and can afford the costs of > > > testing against the S3 service. > > > > > > > > > On Tue, May 18, 2021 at 12:16 PM Wellington Chevreuil < > > > wellington.chevre...@gmail.com> wrote: > > > > > > > Greetings everyone, > > > > > > > > HBASE-24749 has been proposed almost a year ago, introducing a new > > > > StoreFile tracker as a way to allow for any hbase hfile modifications > > to > > > be > > > > safely completed without needing a file system rename. This seems > > pretty > > > > relevant for deployments over S3 file systems, where rename > operations > > > are > > > > not atomic and can have a performance degradation when multiple > > requests > > > > get concurrently
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Thank you, Andrew and Duo, Talking internally with Josh Elser, initial idea was to rebase the feature branch with master (in order to catch with latest commits), then focus on work to have a minimal functioning hbase, in other words, together with the already committed work from HBASE-25391, make sure flush, compactions, splits and merges all can take advantage of the persistent store file manager and complete with no need to rely on renames. These all map to the substasks HBASE-25391, HBASE-25392 and HBASE-25393. Once we could test and validate this works well for our goals, we can then focus on snapshots, bulkloading and tooling. S3 now supports strong consistency, and I heard that they are also > implementing atomic renaming currently, so maybe that's one of the reasons > why the development is silent now.. > Interesting, I had no idea this was being implemented. I know, however, a version of this feature is already available on latest EMR releases (at least from 6.2.0), and AWS team has published their own blog post with their results: https://aws.amazon.com/blogs/big-data/amazon-emr-6-2-0-adds-persistent-hfile-tracking-to-improve-performance-with-hbase-on-amazon-s3/ But I do not think store hfile list in meta is the only solution. It will > cause cyclic dependencies for hbase:meta, and then force us a have a > fallback solution which makes the code a bit ugly. We should try to see if > this could be done with only the FileSystem. > This is indeed a relevant concern. One idea I had mentioned in the original design doc was to track committed/non-committed files through xattr (or tags), which may have its own performance issues as explained by Stephen Wu, but is something that could be attempted. Em qua., 19 de mai. de 2021 às 04:56, 张铎(Duo Zhang) escreveu: > S3 now supports strong consistency, and I heard that they are also > implementing atomic renaming currently, so maybe that's one of the reasons > why the development is silent now... > > For me, I also think deploying hbase on cloud storage is the future, so I > would also like to participate here. > > But I do not think store hfile list in meta is the only solution. It will > cause cyclic dependencies for hbase:meta, and then force us a have a > fallback solution which makes the code a bit ugly. We should try to see if > this could be done with only the FileSystem. > > Thanks. > > Andrew Purtell 于2021年5月19日周三 上午8:04写道: > > > Wellington (and et. al), > > > > S3 is also an important piece of our future production plans. > > Unfortunately, we were unable to assist much with last year's work, on > > account of being sidetracked by more immediate concerns. Fortunately, > this > > renewed interest is timely in that we have an HBase 2 project where, if > > this can land in a 2.5 or a 2.6, it could be an important cost to serve > > optimization, and one we could and would make use of. Therefore I would > > like to restate my employer's interest in this work too. It may just be > > Viraj and myself in the early days. > > > > I'm not sure how best to collaborate. We could review changes from the > > original authors, new changes, and/or divide up the development tasks. We > > can certainly offer our time for testing, and can afford the costs of > > testing against the S3 service. > > > > > > On Tue, May 18, 2021 at 12:16 PM Wellington Chevreuil < > > wellington.chevre...@gmail.com> wrote: > > > > > Greetings everyone, > > > > > > HBASE-24749 has been proposed almost a year ago, introducing a new > > > StoreFile tracker as a way to allow for any hbase hfile modifications > to > > be > > > safely completed without needing a file system rename. This seems > pretty > > > relevant for deployments over S3 file systems, where rename operations > > are > > > not atomic and can have a performance degradation when multiple > requests > > > get concurrently submitted to the same bucket. We had done superficial > > > tests and ycsb runs, where individual renames of files larger than 5GB > > can > > > take a few hundreds of seconds to complete. We also observed impacts in > > > write loads throughput, the bottleneck potentially being the renames. > > > > > > With S3 being an important piece of my employer cloud solution, we > would > > > like to help it move forward. We plan to contribute new patches per the > > > original design/Jira, but we’d also be happy to review changes from the > > > original authors, too. Please let us know if anyone has any concerns, > > > otherwise we’ll start to self-assign issues on HBASE-24749 > > > > > > Wellington > > > > > > > > > -- > > Best regards, > > Andrew > > > > Words like orphans lost among the crosstalk, meaning torn from truth's > > decrepit hands > >- A23, Crosstalk > > >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
S3 now supports strong consistency, and I heard that they are also implementing atomic renaming currently, so maybe that's one of the reasons why the development is silent now... For me, I also think deploying hbase on cloud storage is the future, so I would also like to participate here. But I do not think store hfile list in meta is the only solution. It will cause cyclic dependencies for hbase:meta, and then force us a have a fallback solution which makes the code a bit ugly. We should try to see if this could be done with only the FileSystem. Thanks. Andrew Purtell 于2021年5月19日周三 上午8:04写道: > Wellington (and et. al), > > S3 is also an important piece of our future production plans. > Unfortunately, we were unable to assist much with last year's work, on > account of being sidetracked by more immediate concerns. Fortunately, this > renewed interest is timely in that we have an HBase 2 project where, if > this can land in a 2.5 or a 2.6, it could be an important cost to serve > optimization, and one we could and would make use of. Therefore I would > like to restate my employer's interest in this work too. It may just be > Viraj and myself in the early days. > > I'm not sure how best to collaborate. We could review changes from the > original authors, new changes, and/or divide up the development tasks. We > can certainly offer our time for testing, and can afford the costs of > testing against the S3 service. > > > On Tue, May 18, 2021 at 12:16 PM Wellington Chevreuil < > wellington.chevre...@gmail.com> wrote: > > > Greetings everyone, > > > > HBASE-24749 has been proposed almost a year ago, introducing a new > > StoreFile tracker as a way to allow for any hbase hfile modifications to > be > > safely completed without needing a file system rename. This seems pretty > > relevant for deployments over S3 file systems, where rename operations > are > > not atomic and can have a performance degradation when multiple requests > > get concurrently submitted to the same bucket. We had done superficial > > tests and ycsb runs, where individual renames of files larger than 5GB > can > > take a few hundreds of seconds to complete. We also observed impacts in > > write loads throughput, the bottleneck potentially being the renames. > > > > With S3 being an important piece of my employer cloud solution, we would > > like to help it move forward. We plan to contribute new patches per the > > original design/Jira, but we’d also be happy to review changes from the > > original authors, too. Please let us know if anyone has any concerns, > > otherwise we’ll start to self-assign issues on HBASE-24749 > > > > Wellington > > > > > -- > Best regards, > Andrew > > Words like orphans lost among the crosstalk, meaning torn from truth's > decrepit hands >- A23, Crosstalk >
Re: [DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Wellington (and et. al), S3 is also an important piece of our future production plans. Unfortunately, we were unable to assist much with last year's work, on account of being sidetracked by more immediate concerns. Fortunately, this renewed interest is timely in that we have an HBase 2 project where, if this can land in a 2.5 or a 2.6, it could be an important cost to serve optimization, and one we could and would make use of. Therefore I would like to restate my employer's interest in this work too. It may just be Viraj and myself in the early days. I'm not sure how best to collaborate. We could review changes from the original authors, new changes, and/or divide up the development tasks. We can certainly offer our time for testing, and can afford the costs of testing against the S3 service. On Tue, May 18, 2021 at 12:16 PM Wellington Chevreuil < wellington.chevre...@gmail.com> wrote: > Greetings everyone, > > HBASE-24749 has been proposed almost a year ago, introducing a new > StoreFile tracker as a way to allow for any hbase hfile modifications to be > safely completed without needing a file system rename. This seems pretty > relevant for deployments over S3 file systems, where rename operations are > not atomic and can have a performance degradation when multiple requests > get concurrently submitted to the same bucket. We had done superficial > tests and ycsb runs, where individual renames of files larger than 5GB can > take a few hundreds of seconds to complete. We also observed impacts in > write loads throughput, the bottleneck potentially being the renames. > > With S3 being an important piece of my employer cloud solution, we would > like to help it move forward. We plan to contribute new patches per the > original design/Jira, but we’d also be happy to review changes from the > original authors, too. Please let us know if anyone has any concerns, > otherwise we’ll start to self-assign issues on HBASE-24749 > > Wellington > -- Best regards, Andrew Words like orphans lost among the crosstalk, meaning torn from truth's decrepit hands - A23, Crosstalk
[DISCUSS] Implement and release HBASE-24749 (an hfile tracker that allows for avoiding renames)
Greetings everyone, HBASE-24749 has been proposed almost a year ago, introducing a new StoreFile tracker as a way to allow for any hbase hfile modifications to be safely completed without needing a file system rename. This seems pretty relevant for deployments over S3 file systems, where rename operations are not atomic and can have a performance degradation when multiple requests get concurrently submitted to the same bucket. We had done superficial tests and ycsb runs, where individual renames of files larger than 5GB can take a few hundreds of seconds to complete. We also observed impacts in write loads throughput, the bottleneck potentially being the renames. With S3 being an important piece of my employer cloud solution, we would like to help it move forward. We plan to contribute new patches per the original design/Jira, but we’d also be happy to review changes from the original authors, too. Please let us know if anyone has any concerns, otherwise we’ll start to self-assign issues on HBASE-24749 Wellington