Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
Hi Nick, For brevity, I didn't detail all of the commands I issued. Looking back through my command history, I can confirm that I did explicitly set cache-mode to writeback (and later reset it to forward to try flush-and-evict). Question: how did you determine that your cache-mode was not writeback? I'll do that, just to confirm that this is the problem, then reestablish the cache-mode. Thank you very much for your assistance! -don- -Original Message- From: Nick Fisk [mailto:n...@fisk.me.uk] Sent: 30 April, 2015 10:38 To: Don Doerner; ceph-users@lists.ceph.com Subject: RE: RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal Hi Don, I experienced the same thing a couple of days ago on Hammer. On investigation the cache mode wasn't set to writeback even though I'm sure it accepted the command successfully when I set the pool up. Could you reapply the cache mode writeback command and see if that makes a difference? Nick -Original Message- From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Don Doerner Sent: 30 April 2015 17:57 To: ceph-users@lists.ceph.com Subject: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal All, Synopsis: I can't get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two root level resources a. ctstorage, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. ecstorage, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool ctpool, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool ecpool, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., rados ls shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) - went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full - very unexpected). Using rados df, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities. . I forgot a critical step this time around. . The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). . There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic communications, including email and attachments, sent across its networks filtered through anti virus and spam software programs and retain such messages in order to comply with applicable data security and retention requirements. Quantum is not responsible for the proper and complete transmission of the substance of this communication or for any delay in its receipt. ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
Hi Don, Did you configure target_ dirty_ratio, target_full_ratio and target_max_bytes? K.Mohamed Pakkeer On Thu, Apr 30, 2015 at 10:26 PM, Don Doerner don.doer...@quantum.com wrote: All, Synopsis: I can’t get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two “root” level resources a. “ctstorage”, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. “ecstorage”, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool “ctpool”, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool “ecpool”, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., “rados ls” shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) – went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full – *very* unexpected). Using “rados df”, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities… ·I forgot a critical step this time around. ·The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). ·There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- -- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic communications, including email and attachments, sent across its networks filtered through anti virus and spam software programs and retain such messages in order to comply with applicable data security and retention requirements. Quantum is not responsible for the proper and complete transmission of the substance of this communication or for any delay in its receipt. ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com -- Thanks Regards K.Mohamed Pakkeer Mobile- 0091-8754410114 ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
OK... When I hit the wall, Ceph got pretty unusable; I haven't figured out how to restore it to health. So to do as you suggest, I am going to have to scrape everything into the trash and try again (3rd time's the charm, right?) - so let me get started on that. I will pause before I run the big test that can overflow the cache and consult with you on what specific steps you might recommend. -don- -Original Message- From: Nick Fisk [mailto:n...@fisk.me.uk] Sent: 30 April, 2015 10:58 To: Don Doerner; ceph-users@lists.ceph.com Subject: RE: RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal I'm using Inkscope to monitor my cluster and looking at the pool details I saw that mode was set to none. I'm pretty sure there must be a ceph cmd line to get the option state but I couldn't find anything obvious when I was looking for it. -Original Message- From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Don Doerner Sent: 30 April 2015 18:47 To: Nick Fisk; ceph-users@lists.ceph.com Subject: Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal Hi Nick, For brevity, I didn't detail all of the commands I issued. Looking back through my command history, I can confirm that I did explicitly set cache-mode to writeback (and later reset it to forward to try flush-and-evict). Question: how did you determine that your cache-mode was not writeback? I'll do that, just to confirm that this is the problem, then reestablish the cache- mode. Thank you very much for your assistance! -don- -Original Message- From: Nick Fisk [mailto:n...@fisk.me.uk] Sent: 30 April, 2015 10:38 To: Don Doerner; ceph-users@lists.ceph.com Subject: RE: RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal Hi Don, I experienced the same thing a couple of days ago on Hammer. On investigation the cache mode wasn't set to writeback even though I'm sure it accepted the command successfully when I set the pool up. Could you reapply the cache mode writeback command and see if that makes a difference? Nick -Original Message- From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Don Doerner Sent: 30 April 2015 17:57 To: ceph-users@lists.ceph.com Subject: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal All, Synopsis: I can't get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two root level resources a. ctstorage, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. ecstorage, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool ctpool, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool ecpool, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., rados ls shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) - went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full - very unexpected). Using rados df, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities. . I forgot a critical step this time around. . The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). . There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
I'm using Inkscope to monitor my cluster and looking at the pool details I saw that mode was set to none. I'm pretty sure there must be a ceph cmd line to get the option state but I couldn't find anything obvious when I was looking for it. -Original Message- From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Don Doerner Sent: 30 April 2015 18:47 To: Nick Fisk; ceph-users@lists.ceph.com Subject: Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal Hi Nick, For brevity, I didn't detail all of the commands I issued. Looking back through my command history, I can confirm that I did explicitly set cache-mode to writeback (and later reset it to forward to try flush-and-evict). Question: how did you determine that your cache-mode was not writeback? I'll do that, just to confirm that this is the problem, then reestablish the cache- mode. Thank you very much for your assistance! -don- -Original Message- From: Nick Fisk [mailto:n...@fisk.me.uk] Sent: 30 April, 2015 10:38 To: Don Doerner; ceph-users@lists.ceph.com Subject: RE: RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal Hi Don, I experienced the same thing a couple of days ago on Hammer. On investigation the cache mode wasn't set to writeback even though I'm sure it accepted the command successfully when I set the pool up. Could you reapply the cache mode writeback command and see if that makes a difference? Nick -Original Message- From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Don Doerner Sent: 30 April 2015 17:57 To: ceph-users@lists.ceph.com Subject: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal All, Synopsis: I can't get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two root level resources a. ctstorage, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. ecstorage, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool ctpool, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool ecpool, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., rados ls shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) - went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full - very unexpected). Using rados df, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities. . I forgot a critical step this time around. . The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). . There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic communications, including email and attachments, sent across its networks filtered through anti virus and spam software programs and retain such messages in order to comply with applicable data security and retention requirements. Quantum is not responsible for the proper and complete transmission of the substance of this communication or for any delay in its receipt. ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com ___ ceph-users mailing list ceph-users@lists.ceph.com
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
Hi Don, I experienced the same thing a couple of days ago on Hammer. On investigation the cache mode wasn't set to writeback even though I'm sure it accepted the command successfully when I set the pool up. Could you reapply the cache mode writeback command and see if that makes a difference? Nick -Original Message- From: ceph-users [mailto:ceph-users-boun...@lists.ceph.com] On Behalf Of Don Doerner Sent: 30 April 2015 17:57 To: ceph-users@lists.ceph.com Subject: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Sensitivity: Personal All, Synopsis: I can't get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two root level resources a. ctstorage, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. ecstorage, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool ctpool, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool ecpool, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., rados ls shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) - went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full - very unexpected). Using rados df, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities. . I forgot a critical step this time around. . The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). . There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic communications, including email and attachments, sent across its networks filtered through anti virus and spam software programs and retain such messages in order to comply with applicable data security and retention requirements. Quantum is not responsible for the proper and complete transmission of the substance of this communication or for any delay in its receipt. ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
Hi Mohamed, I did not. But: ·I confirmed that (by default) cache_target_dirty_ratio was set to 0.4 (40%) and cache_target_full_ratio was set to 0.8 (80%). ·I did not set target_max_bytes (I felt that the simple, pure relative sizing controls were sufficient for my experiment). ·I confirmed that (by default) cache_min_flush_age and cache_min_evict_age were set to 0 (so no required delay for either flushing or eviction). Given these settings, I expected to see: ·Flushing begin to happen at 40% of my cache tier size (~200 GB, as it happened), or about 80 GB. Or earlier. ·Eviction begin to happen at 80% of my cache tier size, or about 160 GB. Or earlier. ·Cache tier capacity would only exceed 80% only if the flushing process couldn’t keep up with the ingest process for fairly long periods of time (at the observed ingest rate of ~400 MB/sec, a few hundred seconds). Am I misunderstanding something? Thank you very much for your assistance! -don- From: Mohamed Pakkeer [mailto:mdfakk...@gmail.com] Sent: 30 April, 2015 10:52 To: Don Doerner Cc: ceph-users@lists.ceph.com Subject: Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Hi Don, Did you configure target_ dirty_ratio, target_full_ratio and target_max_bytes? K.Mohamed Pakkeer On Thu, Apr 30, 2015 at 10:26 PM, Don Doerner don.doer...@quantum.commailto:don.doer...@quantum.com wrote: All, Synopsis: I can’t get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two “root” level resources a. “ctstorage”, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. “ecstorage”, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool “ctpool”, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool “ecpool”, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., “rados ls” shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) – went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full – very unexpected). Using “rados df”, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities… •I forgot a critical step this time around. •The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). •There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic communications, including email and attachments, sent across its networks filtered through anti virus and spam software programs and retain such messages in order to comply with applicable data security and retention requirements. Quantum is not responsible for the proper and complete transmission of the substance of this communication or for any delay in its receipt. ___ ceph-users mailing list ceph-users@lists.ceph.commailto:ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.comhttps://urldefense.proofpoint.com/v2/url?u=http-3A__lists.ceph.com_listinfo.cgi_ceph-2Dusers-2Dceph.comd=BQMFaQc=8S5idjlO_n28Ko3lg6lskTMwneSC-WqZ5EBTEEvDlkgr=DAW8QzIBpV_iBddECxqMq8sRPZPQOBqikPmeCBg26bMm=Z-3d5aMnP4pxkHHCAf6pW_kRjxRPDF3dx6MfuVGZDgws=0C6rvtLPHnddUXaVLBff4sszXT6cKSkGmnZag2VVLfke= -- Thanks Regards K.Mohamed Pakkeer Mobile- 0091-8754410114 ___ ceph-users mailing list ceph-users@lists.ceph.com http
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
On Thu, Apr 30, 2015 at 10:57 AM, Nick Fisk n...@fisk.me.uk wrote: I'm using Inkscope to monitor my cluster and looking at the pool details I saw that mode was set to none. I'm pretty sure there must be a ceph cmd line to get the option state but I couldn't find anything obvious when I was looking for it. If you run ceph osd dump then any cache parameters that are set will be output. I just set one up on a vstart cluster and the output includes this line: pool 3 'ctp' replicated size 3 min_size 1 crush_ruleset 0 object_hash rjenkins pg_num 16 pgp_num 16 last_change 10 flags hashpspool,incomplete_clones tier_of 1 cache_mode writeback stripe_width 0 If you dump it with --format=json[-pretty] it will include everything unconditionally and you can inspect that way as well. -Greg ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
Hi Don, You have to provide the target size through target_max_bytes. target_dirty_ratio and target_full_ratio values are based upon target_max_bytes. If you provide target_max bytes as 200 GB and target_dirty_ratio as 0.4, ceph will start the fulshing, when the cache tier has 60GB of dirty objects. Mohamed On Thu, Apr 30, 2015 at 11:56 PM, Don Doerner don.doer...@quantum.com wrote: Hi Mohamed, I did not. But: ·I confirmed that (by default) cache_target_dirty_ratio was set to 0.4 (40%) and cache_target_full_ratio was set to 0.8 (80%). ·I did not set target_max_bytes (I felt that the simple, pure relative sizing controls were sufficient for my experiment). ·I confirmed that (by default) cache_min_flush_age and cache_min_evict_age were set to 0 (so no required delay for either flushing or eviction). Given these settings, I expected to see: ·Flushing begin to happen at 40% of my cache tier size (~200 GB, as it happened), or about 80 GB. Or earlier. ·Eviction begin to happen at 80% of my cache tier size, or about 160 GB. Or earlier. ·Cache tier capacity would only exceed 80% only if the flushing process couldn’t keep up with the ingest process for fairly long periods of time (at the observed ingest rate of ~400 MB/sec, a few hundred seconds). Am I misunderstanding something? Thank you very much for your assistance! -don- *From:* Mohamed Pakkeer [mailto:mdfakk...@gmail.com] *Sent:* 30 April, 2015 10:52 *To:* Don Doerner *Cc:* ceph-users@lists.ceph.com *Subject:* Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Hi Don, Did you configure target_ dirty_ratio, target_full_ratio and target_max_bytes? K.Mohamed Pakkeer On Thu, Apr 30, 2015 at 10:26 PM, Don Doerner don.doer...@quantum.com wrote: All, Synopsis: I can’t get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two “root” level resources a. “ctstorage”, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. “ecstorage”, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool “ctpool”, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool “ecpool”, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., “rados ls” shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) – went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full – *very* unexpected). Using “rados df”, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities… ·I forgot a critical step this time around. ·The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). ·There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- -- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic communications, including email and attachments, sent across its networks filtered through anti virus and spam software programs and retain such messages in order to comply with applicable data security and retention requirements. Quantum is not responsible for the proper and complete transmission of the substance of this communication or for any delay in its receipt. ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com https://urldefense.proofpoint.com
Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict?
Mohamed, Well, that’s interesting… and in direct conflict with what is written in the documentationhttp://ceph.com/docs/master/rados/operations/cache-tiering/ (wherein it describes relative sizing as proportional to the cache pool’s capacity). I am presently reinstalling, so I’ll give that a try. Thanks very much. -don- From: Mohamed Pakkeer [mailto:mdfakk...@gmail.com] Sent: 30 April, 2015 11:45 To: Don Doerner Cc: ceph-users@lists.ceph.com Subject: Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Hi Don, You have to provide the target size through target_max_bytes. target_dirty_ratio and target_full_ratio values are based upon target_max_bytes. If you provide target_max bytes as 200 GB and target_dirty_ratio as 0.4, ceph will start the fulshing, when the cache tier has 60GB of dirty objects. Mohamed On Thu, Apr 30, 2015 at 11:56 PM, Don Doerner don.doer...@quantum.commailto:don.doer...@quantum.com wrote: Hi Mohamed, I did not. But: •I confirmed that (by default) cache_target_dirty_ratio was set to 0.4 (40%) and cache_target_full_ratio was set to 0.8 (80%). •I did not set target_max_bytes (I felt that the simple, pure relative sizing controls were sufficient for my experiment). •I confirmed that (by default) cache_min_flush_age and cache_min_evict_age were set to 0 (so no required delay for either flushing or eviction). Given these settings, I expected to see: •Flushing begin to happen at 40% of my cache tier size (~200 GB, as it happened), or about 80 GB. Or earlier. •Eviction begin to happen at 80% of my cache tier size, or about 160 GB. Or earlier. •Cache tier capacity would only exceed 80% only if the flushing process couldn’t keep up with the ingest process for fairly long periods of time (at the observed ingest rate of ~400 MB/sec, a few hundred seconds). Am I misunderstanding something? Thank you very much for your assistance! -don- From: Mohamed Pakkeer [mailto:mdfakk...@gmail.commailto:mdfakk...@gmail.com] Sent: 30 April, 2015 10:52 To: Don Doerner Cc: ceph-users@lists.ceph.commailto:ceph-users@lists.ceph.com Subject: Re: [ceph-users] RHEL7/HAMMER cache tier doesn't flush or evict? Hi Don, Did you configure target_ dirty_ratio, target_full_ratio and target_max_bytes? K.Mohamed Pakkeer On Thu, Apr 30, 2015 at 10:26 PM, Don Doerner don.doer...@quantum.commailto:don.doer...@quantum.com wrote: All, Synopsis: I can’t get cache tiering to work in HAMMER on RHEL7. Process: 1. Fresh install of HAMMER on RHEL7 went well. 2. Crush map adapted to provide two “root” level resources a. “ctstorage”, to use as a cache tier based on very high-performance, high IOPS SSD (intrinsic journal). 2 OSDs. b. “ecstorage”, to use as an erasure-coded poolbased on low-performance, cost effective storage (extrinsic journal). 12 OSDs. 3. Established a pool “ctpool”, 32 PGs on ctstorage (pool size = min_size = 1). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 4. Established a pool “ecpool”, 256 PGs on ecstorage (5+3 profile). Ran a quick RADOS bench test, all worked as expected. Cleaned up. 5. Ensured that both pools were empty (i.e., “rados ls” shows no objects) 6. Put the cache tier on the erasure coded storage (one Bloom hit set, interval 900 seconds), set up the overlay. Used defaults for flushing and eviction. No errors. 7. Started a 3600-second write test to ecpool. Objects piled up in ctpool (as expected) – went past the 40% mark (as expected), then past the 80% mark (unexpected), then ran into the wall (95% full – very unexpected). Using “rados df”, I can see that the cache tier is full (duh!) but not one single object lives in the ecpool. Nothing was ever flushed, nothing was ever evicted. Thought I might be misreading that, so I went back to SAR data that I captured during the test: the SSDs were the only [ceph] devices that sustained I/O. I based this experiment on another (much more successful) experiment that I performed using GIANT (.1) on RHEL7 a couple of weeks ago (wherein I used RAM as a cache tier); that all worked. It seems there are at least three possibilities… •I forgot a critical step this time around. •The steps needed for a cache tier in HAMMER are different than the steps needed in GIANT (and different than the documentation online). •There is a problem with HAMMER in the area of cache tier. Has anyone successfully deployed cache-tiering in HAMMER? Did you have to do anything unusual? Do you see any steps that I missed? Regards, -don- The information contained in this transmission may be confidential. Any disclosure, copying, or further distribution of confidential information is not permitted unless such privilege is explicitly granted in writing by Quantum. Quantum reserves the right to have electronic