Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 15/09/2016 17:23, Alex Bligh wrote: > Paolo, > >> On 15 Sep 2016, at 15:07, Paolo Bonzini wrote: >> >> I don't think QEMU forbids multiple clients to the single server, and >> guarantees consistency as long as there is no overlap between writes and >> reads. These are the same guarantees you have for multiple commands on >> a single connection. >> >> In other words, from the POV of QEMU there's no difference whether >> multiple commands come from one or more connections. > > This isn't really about ordering, it's about cache coherency > and persisting things to disk. > > What you say is correct as far as it goes in terms of ordering. However > consider the scenario with read and writes on two channels as follows > of the same block: > > Channel1 Channel2 > > R Block read, and cached in user space in > channel 1's cache > Reply sent > > W New value written, channel 2's cache updated > channel 1's cache not > > R Value returned from channel 1's cache. > > > In the above scenario, there is a problem if the server(s) handling the > two channels each use a read cache which is not coherent between the > two channels. An example would be a read-through cache on a server that > did fork() and shared no state between connections. qemu-nbd does not fork(), so there is no coherency issue if W has replied. However, if W hasn't replied, channel1 can get garbage. Typically the VM will be the one during writes, everyone else must be ready to handle whatever mess the VM throws at them. Paolo > Similarly, if there is a write on channel 1 that has completed, and > the flush goes to channel 2, it may not (if state is not shared) guarantee > that the write on channel 1 (which has completed) is persisted to non-volatile > media. Obviously if the 'state' is OS block cache/buffers/whatever, it > will, but if it's (e.g.) a user-space per process write-through cache, > it won't. > > I don't know whether qemu-nbd is likely to suffer from either of these. It can't happen. On the other hand, channel1 must be ready to handle garbage, it's illegal. signature.asc Description: OpenPGP digital signature
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 09/15/2016 11:27 AM, Wouter Verhelst wrote: > On Thu, Sep 15, 2016 at 05:08:21PM +0100, Alex Bligh wrote: >> Wouter, >> >>> The server can always refuse to allow multiple connections. >> >> Sure, but it would be neater to warn the client of that at negotiation >> stage (it would only be one flag, e.g. 'multiple connections >> unsafe'). > > I suppose that's not a bad idea. Seems like it may need to be a per-export flag, rather than a global flag (as a given server may be able to serve multiple types of files, where the safety depends on the type of file being served). -- Eric Blake eblake redhat com+1-919-301-3266 Libvirt virtualization library http://libvirt.org signature.asc Description: OpenPGP digital signature
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Wouter, > On 15 Sep 2016, at 17:27, Wouter Verhelst wrote: > > On Thu, Sep 15, 2016 at 05:08:21PM +0100, Alex Bligh wrote: >> Wouter, >> >>> The server can always refuse to allow multiple connections. >> >> Sure, but it would be neater to warn the client of that at negotiation >> stage (it would only be one flag, e.g. 'multiple connections >> unsafe'). > > I suppose that's not a bad idea. Good. > [...] >>> I was thinking of changing the spec as follows: >>> >>> diff --git a/doc/proto.md b/doc/proto.md >>> index 217f57e..cb099e2 100644 >>> --- a/doc/proto.md >>> +++ b/doc/proto.md >>> @@ -308,6 +308,23 @@ specification, the >>> [kernel >>> documentation](https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt) >>> may be useful. >>> >>> +For performance reasons, clients MAY open multiple connections to the >>> +same server. To support such clients, servers SHOULD ensure that at >>> +least one of the following conditions hold: >>> + >>> +* Flush commands are processed for ALL connections. That is, when an >>> + `NBD_CMD_WRITE` is processed on one connection, and then an >>> + `NBD_CMD_FLUSH` is processed on another connection, the data of the >>> + `NBD_CMD_WRITE` on the first connection MUST reach permanent storage >>> + before the reply of the `NBD_CMD_FLUSH` is sent. >>> +* The server allows `NBD_CMD_WRITE` and `NBD_CMD_FLUSH` on at most one >>> + connection >>> +* Multiple connections are not allowed >>> + >>> +In addition, clients using multiple connections SHOULD NOT send >>> +`NBD_CMD_FLUSH` if an `NBD_CMD_WRITE` for which they care in relation to >>> +the flush has not been replied to yet. >>> + >> >> I don't think that should be a mandatory behaviour. > > Which part of it? I've read it again :-) The wording was slightly contorted. I think what I mean is that if you don't support flush at all, that's another option. The final paragraph I am not sure is right, as that's not what the kernel currently does. If we are going to suggest a change in our main client's behaviour, should we not just request that flush is done on all channels? >> For once, it would >> be reasonably easy on gonbdserver but a PITA on the reference server. >> You'd need to put in IPC between each of the forked processes OR rely >> on fdatasync() only - I'm not sure that would necessarily work >> safely with (e.g.) the 'treefiles' / COW options. >> >> I think better would be to say that the server MUST either >> >> * Not support NBD_CMD_FLUSH at all > > I think we should discourage not supporting FLUSH, rather than > suggesting it. Sure, but some backends just don't support flush. For them, this aspect at least is not a problem. >> * Support NBD_CMD_FLUSH across channels (as you set out above), or >> * Indicate that it does not support multiple channels. > > You dropped the one with no writes. I said "at most" there for a reason. > Originally I was going to say "if the server is read-only", but then > thought that it could work to do the "at most" thing. After having given > that some more thought, I now realize that if you write, you need to > flush across to other channels, regardless of whether they write too, so > that bit of it is moot now anyway. > > Still, a server which exports read-only should still be safe for > multiple connections, even if there is no cache coherency (since > presumably nothing's going to change anyway). Yes >> Actually I think this is a problem anyway. A simpler failure case is >> one where (by chance) one channel gets the writes, and one channel >> gets the flushes. The flush reply is delayed beyond the replies to >> unconnected writes (on the other channel) and hence the kernel thinks >> replied-to writes have been persisted when they have not been. > > Yes, that is another example of essentially the same problem. Yeah, I was just trying to simplify things. >> The only way to fix that (as far as I can see) without changing flush >> semantics is for the block layer to issue flush requests on each >> channel when flushing on one channel. > > Christoph just said that that doesn't (currently) happen; I don't know > whether the kernel currently already (optimistically) sends out flush > requests before the writes that it expects to hit permanent storage have > finished, but if it doesn't do that, then there is no problem and my > suggested bit of spec would be okay. > > If there are good reasons to do so, however, we do indeed have a problem > and something else is necessary. I don't think flushing across all > connections is the best solution, though. Well, the way I look at it is that we have a proposed change in client behaviour (multiple channels) which causes problems at least with flush and also (I think) with cache coherency (see other email). We should either not make that change, or ensure other changes are added which mitigate these issues. Flush is actually the obvious one. Cache coherency is far more subtle (though possibly fixable by s
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 05:08:21PM +0100, Alex Bligh wrote: > Wouter, > > > The server can always refuse to allow multiple connections. > > Sure, but it would be neater to warn the client of that at negotiation > stage (it would only be one flag, e.g. 'multiple connections > unsafe'). I suppose that's not a bad idea. [...] > > I was thinking of changing the spec as follows: > > > > diff --git a/doc/proto.md b/doc/proto.md > > index 217f57e..cb099e2 100644 > > --- a/doc/proto.md > > +++ b/doc/proto.md > > @@ -308,6 +308,23 @@ specification, the > > [kernel > > documentation](https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt) > > may be useful. > > > > +For performance reasons, clients MAY open multiple connections to the > > +same server. To support such clients, servers SHOULD ensure that at > > +least one of the following conditions hold: > > + > > +* Flush commands are processed for ALL connections. That is, when an > > + `NBD_CMD_WRITE` is processed on one connection, and then an > > + `NBD_CMD_FLUSH` is processed on another connection, the data of the > > + `NBD_CMD_WRITE` on the first connection MUST reach permanent storage > > + before the reply of the `NBD_CMD_FLUSH` is sent. > > +* The server allows `NBD_CMD_WRITE` and `NBD_CMD_FLUSH` on at most one > > + connection > > +* Multiple connections are not allowed > > + > > +In addition, clients using multiple connections SHOULD NOT send > > +`NBD_CMD_FLUSH` if an `NBD_CMD_WRITE` for which they care in relation to > > +the flush has not been replied to yet. > > + > > I don't think that should be a mandatory behaviour. Which part of it? > For once, it would > be reasonably easy on gonbdserver but a PITA on the reference server. > You'd need to put in IPC between each of the forked processes OR rely > on fdatasync() only - I'm not sure that would necessarily work > safely with (e.g.) the 'treefiles' / COW options. > > I think better would be to say that the server MUST either > > * Not support NBD_CMD_FLUSH at all I think we should discourage not supporting FLUSH, rather than suggesting it. > * Support NBD_CMD_FLUSH across channels (as you set out above), or > * Indicate that it does not support multiple channels. You dropped the one with no writes. I said "at most" there for a reason. Originally I was going to say "if the server is read-only", but then thought that it could work to do the "at most" thing. After having given that some more thought, I now realize that if you write, you need to flush across to other channels, regardless of whether they write too, so that bit of it is moot now anyway. Still, a server which exports read-only should still be safe for multiple connections, even if there is no cache coherency (since presumably nothing's going to change anyway). [...] > > The latter bit (on the client side) is because even if your backend has > > no cache coherency issues, TCP does not guarantee ordering between > > multiple connections. I don't know if the above is in line with what > > blk-mq does, but consider the following scenario: > > > > - A client sends two writes to the server, followed (immediately) by a > > flush, where at least the second write and the flush are not sent over > > the same connection. > > - The first write is a small one, and it is handled almost immediately. > > - The second write takes a little longer, so the flush is handled > > earlier than the second write > > - The network packet containing the flush reply gets lost for whatever > > reason, so the client doesn't get it, and we fall into TCP > > retransmits. > > - The second write finishes, and its reply header does not get lost > > - After the second write reply reaches the client, the TCP retransmits > > for the flush reply are handled. > > > > In the above scenario, the flush reply arrives on the client side after > > a write reply which it did not cover; so the client will (incorrectly) > > assume that the write has reached permanent storage when in fact it may > > not have done so yet. > > > > If the kernel does not care about the ordering of the two writes versus > > the flush, then there is no problem. I don't know how blk-mq works in > > that context, but if the above is a likely scenario, we may have to > > reconsider adding blk-mq to nbd. > > Actually I think this is a problem anyway. A simpler failure case is > one where (by chance) one channel gets the writes, and one channel > gets the flushes. The flush reply is delayed beyond the replies to > unconnected writes (on the other channel) and hence the kernel thinks > replied-to writes have been persisted when they have not been. Yes, that is another example of essentially the same problem. > The only way to fix that (as far as I can see) without changing flush > semantics is for the block layer to issue flush requests on each > channel when flushing on one channel. Christoph just said that that doesn't (currently) happen; I don't know whether
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Wouter, > The server can always refuse to allow multiple connections. Sure, but it would be neater to warn the client of that at negotiation stage (it would only be one flag, e.g. 'multiple connections unsafe'). That way the connection won't fail with a cryptic EBUSY or whatever, but will just negotiate a single connection. > I was thinking of changing the spec as follows: > > diff --git a/doc/proto.md b/doc/proto.md > index 217f57e..cb099e2 100644 > --- a/doc/proto.md > +++ b/doc/proto.md > @@ -308,6 +308,23 @@ specification, the > [kernel > documentation](https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt) > may be useful. > > +For performance reasons, clients MAY open multiple connections to the > +same server. To support such clients, servers SHOULD ensure that at > +least one of the following conditions hold: > + > +* Flush commands are processed for ALL connections. That is, when an > + `NBD_CMD_WRITE` is processed on one connection, and then an > + `NBD_CMD_FLUSH` is processed on another connection, the data of the > + `NBD_CMD_WRITE` on the first connection MUST reach permanent storage > + before the reply of the `NBD_CMD_FLUSH` is sent. > +* The server allows `NBD_CMD_WRITE` and `NBD_CMD_FLUSH` on at most one > + connection > +* Multiple connections are not allowed > + > +In addition, clients using multiple connections SHOULD NOT send > +`NBD_CMD_FLUSH` if an `NBD_CMD_WRITE` for which they care in relation to > +the flush has not been replied to yet. > + I don't think that should be a mandatory behaviour. For once, it would be reasonably easy on gonbdserver but a PITA on the reference server. You'd need to put in IPC between each of the forked processes OR rely on fdatasync() only - I'm not sure that would necessarily work safely with (e.g.) the 'treefiles' / COW options. I think better would be to say that the server MUST either * Not support NBD_CMD_FLUSH at all * Support NBD_CMD_FLUSH across channels (as you set out above), or * Indicate that it does not support multiple channels. > Request message > > The request message, sent by the client, looks as follows: > > The latter bit (on the client side) is because even if your backend has > no cache coherency issues, TCP does not guarantee ordering between > multiple connections. I don't know if the above is in line with what > blk-mq does, but consider the following scenario: > > - A client sends two writes to the server, followed (immediately) by a > flush, where at least the second write and the flush are not sent over > the same connection. > - The first write is a small one, and it is handled almost immediately. > - The second write takes a little longer, so the flush is handled > earlier than the second write > - The network packet containing the flush reply gets lost for whatever > reason, so the client doesn't get it, and we fall into TCP > retransmits. > - The second write finishes, and its reply header does not get lost > - After the second write reply reaches the client, the TCP retransmits > for the flush reply are handled. > > In the above scenario, the flush reply arrives on the client side after > a write reply which it did not cover; so the client will (incorrectly) > assume that the write has reached permanent storage when in fact it may > not have done so yet. > > If the kernel does not care about the ordering of the two writes versus > the flush, then there is no problem. I don't know how blk-mq works in > that context, but if the above is a likely scenario, we may have to > reconsider adding blk-mq to nbd. Actually I think this is a problem anyway. A simpler failure case is one where (by chance) one channel gets the writes, and one channel gets the flushes. The flush reply is delayed beyond the replies to unconnected writes (on the other channel) and hence the kernel thinks replied-to writes have been persisted when they have not been. The only way to fix that (as far as I can see) without changing flush semantics is for the block layer to issue flush requests on each channel when flushing on one channel. This, incidentally, would resolve every other issue! -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Eric, > I doubt that qemu-nbd would ever want to support the situation with more > than one client connection writing to the same image at the same time; > the implications of sorting out data consistency between multiple > writers is rather complex and not worth coding into qemu. So I think > qemu would probably prefer to just prohibit the multiple writer > situation. Yeah, I was thinking about a 'no multiple connection' flag. > And while multiple readers with no writer should be fine, > I'm not even sure if multiple readers plus one writer can always be made > to appear sane (if there is no coordination between the different > connections, on an image where the writer changes AA to BA then flushes > then changes to BB, it is still feasible that a reader could see AB > (pre-flush state of the first sector, post-flush changes to the second > sector, even though the writer never flushed that particular content to > disk). Agree -- Alex Bligh signature.asc Description: Message signed with OpenPGP using GPGMail
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Paolo, > On 15 Sep 2016, at 15:07, Paolo Bonzini wrote: > > I don't think QEMU forbids multiple clients to the single server, and > guarantees consistency as long as there is no overlap between writes and > reads. These are the same guarantees you have for multiple commands on > a single connection. > > In other words, from the POV of QEMU there's no difference whether > multiple commands come from one or more connections. This isn't really about ordering, it's about cache coherency and persisting things to disk. What you say is correct as far as it goes in terms of ordering. However consider the scenario with read and writes on two channels as follows of the same block: Channel1 Channel2 R Block read, and cached in user space in channel 1's cache Reply sent W New value written, channel 2's cache updated channel 1's cache not R Value returned from channel 1's cache. In the above scenario, there is a problem if the server(s) handling the two channels each use a read cache which is not coherent between the two channels. An example would be a read-through cache on a server that did fork() and shared no state between connections. Similarly, if there is a write on channel 1 that has completed, and the flush goes to channel 2, it may not (if state is not shared) guarantee that the write on channel 1 (which has completed) is persisted to non-volatile media. Obviously if the 'state' is OS block cache/buffers/whatever, it will, but if it's (e.g.) a user-space per process write-through cache, it won't. I don't know whether qemu-nbd is likely to suffer from either of these. -- Alex Bligh signature.asc Description: Message signed with OpenPGP using GPGMail
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Josef, > On 15 Sep 2016, at 14:57, Josef Bacik wrote: > > This isn't an NBD problem, this is an application problem. The application > must wait for all writes it cares about _before_ issuing a flush. This is > the same as for normal storage as it is for NBD. It is not NBD's > responsibility to maintain coherency between multiple requests across > connections, just simply to act on and respond to requests. > > I think changing the specification to indicate that this is the case for > multiple connections is a good thing, to keep NBD servers from doing weird > things like sending different connections to the same export to different > backing stores without some sort of synchronization. It should definitely be > explicitly stated somewhere that NBD does not provide any ordering guarantees > and that is up to the application. Thanks, I don't think that's correct. The block stack issues a flush to mean (precisely) "do not reply to this until all preceding writes THAT HAVE BEEN REPLIED TO have been persisted to non-volatile storage". The danger is with multiple connections (where apparently only one flush is sent - let's say down connection 1) that not al the writes that have been replied to on connection 2 have been persisted to non-volatile storage. Only the ones on connection 1 have been persisted (this is assuming the nbd server doesn't 'link' in some way the connections). There's nothing the 'application' (here meaning the kernel or higher level) can do to mitigate this. Sure it can wait for all the replies, but this doesn't guarantee the writes have been persisted to non-volatile storage, precisely because writes may return prior to this. -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 15/09/2016 15:34, Eric Blake wrote: > On 09/15/2016 06:09 AM, Alex Bligh wrote: >> >> I also wonder whether any servers that can do caching per >> connection will always share a consistent cache between >> connections. The one I'm worried about in particular here >> is qemu-nbd - Eric Blake CC'd. >> > > I doubt that qemu-nbd would ever want to support the situation with more > than one client connection writing to the same image at the same time; > the implications of sorting out data consistency between multiple > writers is rather complex and not worth coding into qemu. So I think > qemu would probably prefer to just prohibit the multiple writer > situation. And while multiple readers with no writer should be fine, > I'm not even sure if multiple readers plus one writer can always be made > to appear sane (if there is no coordination between the different > connections, on an image where the writer changes AA to BA then flushes > then changes to BB, it is still feasible that a reader could see AB > (pre-flush state of the first sector, post-flush changes to the second > sector, even though the writer never flushed that particular content to > disk). > > But Paolo Bonzini (cc'd) may have more insight on qemu's NBD server and > what it supports (or forbids) in the way of multiple clients to a single > server. I don't think QEMU forbids multiple clients to the single server, and guarantees consistency as long as there is no overlap between writes and reads. These are the same guarantees you have for multiple commands on a single connection. In other words, from the POV of QEMU there's no difference whether multiple commands come from one or more connections. Paolo >> A more general point is that with multiple queues requests >> may be processed in a different order even by those servers that >> currently process the requests in strict order, or in something >> similar to strict order. The server is permitted by the spec >> (save as mandated by NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA) to >> process commands out of order anyway, but I suspect this has >> to date been little tested. > > qemu-nbd is definitely capable of serving reads and writes out-of-order > to a single connection client; but that's different than the case with > multiple connections. > signature.asc Description: OpenPGP digital signature
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 09/15/2016 09:17 AM, Wouter Verhelst wrote: On Thu, Sep 15, 2016 at 01:44:29PM +0100, Alex Bligh wrote: On 15 Sep 2016, at 13:41, Christoph Hellwig wrote: On Thu, Sep 15, 2016 at 01:39:11PM +0100, Alex Bligh wrote: That's probably right in the case of file-based back ends that are running on a Linux OS. But gonbdserver for instance supports (e.g.) Ceph based backends, where each connection might be talking to a completely separate ceph node, and there may be no cache consistency between connections. Yes, if you don't have a cache coherent backend you are generally screwed with a multiqueue protocol. I wonder if the ability to support multiqueue should be visible in the negotiation stage. That would allow the client to refuse to select multiqueue where it isn't safe. The server can always refuse to allow multiple connections. I was thinking of changing the spec as follows: diff --git a/doc/proto.md b/doc/proto.md index 217f57e..cb099e2 100644 --- a/doc/proto.md +++ b/doc/proto.md @@ -308,6 +308,23 @@ specification, the [kernel documentation](https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt) may be useful. +For performance reasons, clients MAY open multiple connections to the +same server. To support such clients, servers SHOULD ensure that at +least one of the following conditions hold: + +* Flush commands are processed for ALL connections. That is, when an + `NBD_CMD_WRITE` is processed on one connection, and then an + `NBD_CMD_FLUSH` is processed on another connection, the data of the + `NBD_CMD_WRITE` on the first connection MUST reach permanent storage + before the reply of the `NBD_CMD_FLUSH` is sent. +* The server allows `NBD_CMD_WRITE` and `NBD_CMD_FLUSH` on at most one + connection +* Multiple connections are not allowed + +In addition, clients using multiple connections SHOULD NOT send +`NBD_CMD_FLUSH` if an `NBD_CMD_WRITE` for which they care in relation to +the flush has not been replied to yet. + Request message The request message, sent by the client, looks as follows: The latter bit (on the client side) is because even if your backend has no cache coherency issues, TCP does not guarantee ordering between multiple connections. I don't know if the above is in line with what blk-mq does, but consider the following scenario: - A client sends two writes to the server, followed (immediately) by a flush, where at least the second write and the flush are not sent over the same connection. - The first write is a small one, and it is handled almost immediately. - The second write takes a little longer, so the flush is handled earlier than the second write - The network packet containing the flush reply gets lost for whatever reason, so the client doesn't get it, and we fall into TCP retransmits. - The second write finishes, and its reply header does not get lost - After the second write reply reaches the client, the TCP retransmits for the flush reply are handled. In the above scenario, the flush reply arrives on the client side after a write reply which it did not cover; so the client will (incorrectly) assume that the write has reached permanent storage when in fact it may not have done so yet. This isn't an NBD problem, this is an application problem. The application must wait for all writes it cares about _before_ issuing a flush. This is the same as for normal storage as it is for NBD. It is not NBD's responsibility to maintain coherency between multiple requests across connections, just simply to act on and respond to requests. I think changing the specification to indicate that this is the case for multiple connections is a good thing, to keep NBD servers from doing weird things like sending different connections to the same export to different backing stores without some sort of synchronization. It should definitely be explicitly stated somewhere that NBD does not provide any ordering guarantees and that is up to the application. Thanks, Josef -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 09/15/2016 06:09 AM, Alex Bligh wrote: > > I also wonder whether any servers that can do caching per > connection will always share a consistent cache between > connections. The one I'm worried about in particular here > is qemu-nbd - Eric Blake CC'd. > I doubt that qemu-nbd would ever want to support the situation with more than one client connection writing to the same image at the same time; the implications of sorting out data consistency between multiple writers is rather complex and not worth coding into qemu. So I think qemu would probably prefer to just prohibit the multiple writer situation. And while multiple readers with no writer should be fine, I'm not even sure if multiple readers plus one writer can always be made to appear sane (if there is no coordination between the different connections, on an image where the writer changes AA to BA then flushes then changes to BB, it is still feasible that a reader could see AB (pre-flush state of the first sector, post-flush changes to the second sector, even though the writer never flushed that particular content to disk). But Paolo Bonzini (cc'd) may have more insight on qemu's NBD server and what it supports (or forbids) in the way of multiple clients to a single server. > A more general point is that with multiple queues requests > may be processed in a different order even by those servers that > currently process the requests in strict order, or in something > similar to strict order. The server is permitted by the spec > (save as mandated by NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA) to > process commands out of order anyway, but I suspect this has > to date been little tested. qemu-nbd is definitely capable of serving reads and writes out-of-order to a single connection client; but that's different than the case with multiple connections. -- Eric Blake eblake redhat com+1-919-301-3266 Libvirt virtualization library http://libvirt.org signature.asc Description: OpenPGP digital signature
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 01:44:29PM +0100, Alex Bligh wrote: > > > On 15 Sep 2016, at 13:41, Christoph Hellwig wrote: > > > > On Thu, Sep 15, 2016 at 01:39:11PM +0100, Alex Bligh wrote: > >> That's probably right in the case of file-based back ends that > >> are running on a Linux OS. But gonbdserver for instance supports > >> (e.g.) Ceph based backends, where each connection might be talking > >> to a completely separate ceph node, and there may be no cache > >> consistency between connections. > > > > Yes, if you don't have a cache coherent backend you are generally > > screwed with a multiqueue protocol. > > I wonder if the ability to support multiqueue should be visible > in the negotiation stage. That would allow the client to refuse > to select multiqueue where it isn't safe. The server can always refuse to allow multiple connections. I was thinking of changing the spec as follows: diff --git a/doc/proto.md b/doc/proto.md index 217f57e..cb099e2 100644 --- a/doc/proto.md +++ b/doc/proto.md @@ -308,6 +308,23 @@ specification, the [kernel documentation](https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt) may be useful. +For performance reasons, clients MAY open multiple connections to the +same server. To support such clients, servers SHOULD ensure that at +least one of the following conditions hold: + +* Flush commands are processed for ALL connections. That is, when an + `NBD_CMD_WRITE` is processed on one connection, and then an + `NBD_CMD_FLUSH` is processed on another connection, the data of the + `NBD_CMD_WRITE` on the first connection MUST reach permanent storage + before the reply of the `NBD_CMD_FLUSH` is sent. +* The server allows `NBD_CMD_WRITE` and `NBD_CMD_FLUSH` on at most one + connection +* Multiple connections are not allowed + +In addition, clients using multiple connections SHOULD NOT send +`NBD_CMD_FLUSH` if an `NBD_CMD_WRITE` for which they care in relation to +the flush has not been replied to yet. + Request message The request message, sent by the client, looks as follows: The latter bit (on the client side) is because even if your backend has no cache coherency issues, TCP does not guarantee ordering between multiple connections. I don't know if the above is in line with what blk-mq does, but consider the following scenario: - A client sends two writes to the server, followed (immediately) by a flush, where at least the second write and the flush are not sent over the same connection. - The first write is a small one, and it is handled almost immediately. - The second write takes a little longer, so the flush is handled earlier than the second write - The network packet containing the flush reply gets lost for whatever reason, so the client doesn't get it, and we fall into TCP retransmits. - The second write finishes, and its reply header does not get lost - After the second write reply reaches the client, the TCP retransmits for the flush reply are handled. In the above scenario, the flush reply arrives on the client side after a write reply which it did not cover; so the client will (incorrectly) assume that the write has reached permanent storage when in fact it may not have done so yet. If the kernel does not care about the ordering of the two writes versus the flush, then there is no problem. I don't know how blk-mq works in that context, but if the above is a likely scenario, we may have to reconsider adding blk-mq to nbd. -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 13:41, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 01:39:11PM +0100, Alex Bligh wrote: >> That's probably right in the case of file-based back ends that >> are running on a Linux OS. But gonbdserver for instance supports >> (e.g.) Ceph based backends, where each connection might be talking >> to a completely separate ceph node, and there may be no cache >> consistency between connections. > > Yes, if you don't have a cache coherent backend you are generally > screwed with a multiqueue protocol. I wonder if the ability to support multiqueue should be visible in the negotiation stage. That would allow the client to refuse to select multiqueue where it isn't safe. Wouter? -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 01:39:11PM +0100, Alex Bligh wrote: > That's probably right in the case of file-based back ends that > are running on a Linux OS. But gonbdserver for instance supports > (e.g.) Ceph based backends, where each connection might be talking > to a completely separate ceph node, and there may be no cache > consistency between connections. Yes, if you don't have a cache coherent backend you are generally screwed with a multiqueue protocol. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 13:36, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 01:33:20PM +0100, Alex Bligh wrote: >> At an implementation level that is going to be a little difficult >> for some NBD servers, e.g. ones that fork() a different process per >> connection. There is in general no IPC to speak of between server >> instances. Such servers would thus be unsafe with more than one >> connection if FLUSH is in use. >> >> I believe such servers include the reference server where there is >> process per connection (albeit possibly with several threads). >> >> Even single process servers (including mine - gonbdserver) would >> require logic to pair up multiple connections to the same >> device. > > Why? If you only send the completion after your I/O syscall returned > your are fine if fsync comes from a difference process, no matter > if you're using direct or buffered I/O underneath. That's probably right in the case of file-based back ends that are running on a Linux OS. But gonbdserver for instance supports (e.g.) Ceph based backends, where each connection might be talking to a completely separate ceph node, and there may be no cache consistency between connections. -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 01:33:20PM +0100, Alex Bligh wrote: > At an implementation level that is going to be a little difficult > for some NBD servers, e.g. ones that fork() a different process per > connection. There is in general no IPC to speak of between server > instances. Such servers would thus be unsafe with more than one > connection if FLUSH is in use. > > I believe such servers include the reference server where there is > process per connection (albeit possibly with several threads). > > Even single process servers (including mine - gonbdserver) would > require logic to pair up multiple connections to the same > device. Why? If you only send the completion after your I/O syscall returned your are fine if fsync comes from a difference process, no matter if you're using direct or buffered I/O underneath. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 13:23, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 02:21:20PM +0200, Wouter Verhelst wrote: >> Right. So do I understand you correctly that blk-mq currently doesn't >> look at multiple queues, and just assumes that if a FLUSH is sent over >> any one of the queues, it applies to all queues? > > Yes. The same is true at the protocol level for NVMe or SCSI transports > that can make use of multiple queues. At an implementation level that is going to be a little difficult for some NBD servers, e.g. ones that fork() a different process per connection. There is in general no IPC to speak of between server instances. Such servers would thus be unsafe with more than one connection if FLUSH is in use. I believe such servers include the reference server where there is process per connection (albeit possibly with several threads). Even single process servers (including mine - gonbdserver) would require logic to pair up multiple connections to the same device. I suspect ALL nbd servers would require a change. So I think we should think carefully before introducing this protocol change. It might be easier to make the client code change work around this issue (somehow). -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 13:18, Christoph Hellwig wrote: > > Yes, please do that. A "barrier" implies draining of the queue. Done -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 02:26:31PM +0200, Wouter Verhelst wrote: > Yes. I think the kernel nbd driver should probably filter out FUA on > READ. It has no meaning in the case of nbd, and whatever expectations > the kernel may have cannot be provided for by nbd anyway. The kernel never sets FUA on reads - I just pointed out that other protocols have defined (although horrible) semantics for it. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 05:20:08AM -0700, Christoph Hellwig wrote: > On Thu, Sep 15, 2016 at 02:01:59PM +0200, Wouter Verhelst wrote: > > Yes. There was some discussion on that part, and we decided that setting > > the flag doesn't hurt, but the spec also clarifies that using it on READ > > does nothing, semantically. > > > > > > The problem is that there are clients in the wild which do set it on > > READ, so it's just a matter of "be liberal in what you accept". > > Note that FUA on READ in SCSI and NVMe does have a meaning - it > requires you to bypass any sort of cache on the target. I think it's an > wrong defintion because it mandates implementation details that aren't > observable by the initiator, but it's still the spec wording and nbd > diverges from it. That's not nessecarily a bad thing, but a caveat to > look out for. Yes. I think the kernel nbd driver should probably filter out FUA on READ. It has no meaning in the case of nbd, and whatever expectations the kernel may have cannot be provided for by nbd anyway. -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 02:21:20PM +0200, Wouter Verhelst wrote: > Right. So do I understand you correctly that blk-mq currently doesn't > look at multiple queues, and just assumes that if a FLUSH is sent over > any one of the queues, it applies to all queues? Yes. The same is true at the protocol level for NVMe or SCSI transports that can make use of multiple queues. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 05:01:25AM -0700, Christoph Hellwig wrote: > On Thu, Sep 15, 2016 at 01:55:14PM +0200, Wouter Verhelst wrote: > > If that's not a write barrier, then I was using the wrong terminology > > (and offer my apologies for the confusion). > > It's not a write barrier - a write barrier was command that ensured that [...] Okay, I'll update the spec to remove that term then. > > The point still remains that "X was sent before Y" is difficult to > > determine on the client side if X was sent over a different TCP channel > > than Y, because a packet might be dropped (requiring a retransmit) for > > X, and similar things. If blk-mq can deal with that, we're good and > > nothing further needs to be done. If not, this should be evaluated by > > someone more familiar with the internals of the kernel block subsystem > > than me. > > The important bit in all the existing protocols, and which Linux relies > on is that any write the Linux block layer got a completion for earlier > needs to be flushed out to non-volatile storage when a FLUSH command is > set. Anything that still is in flight does not matter. Which for > NBD means anything that you already replies to need to be flushed. > > Or to say it more practicly - in the nbd server you simply need to > call fdatasync on the backing device or file whenever you get a FLUSH > requires, and it will do the right thing. Right. So do I understand you correctly that blk-mq currently doesn't look at multiple queues, and just assumes that if a FLUSH is sent over any one of the queues, it applies to all queues? If so, I'll update the spec to clarify that servers should ensure this holds in the case of multiple connections, or not allow writes, or not allow multiple connections. -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 01:11:24PM +0100, Alex Bligh wrote: > > NBD_CMD_FLUSH (3) > > > > A flush request; a write barrier. > > I can see that's potentially confusing as isn't meant to mean 'an old-style > linux kernel block device write barrier'. I think in general terms it > probably is some form of barrier, but I see no problem in deleting the > words "a write barrier" from the spec text if only to make it > clearer. Yes, please do that. A "barrier" implies draining of the queue. > However, I think the description of the command itself: > > > The server MUST NOT send a successful reply header for this request before > > all write requests for which a reply has already been sent to the client > > have reached permanent storage (using fsync() or similar). > > and the ordering section I pointed you to before, were both correct, yes? Yes, this seems correct. > actually fdatasync() technically does more than is necessary, as it > will also flush commands that have been processed, but for which no > reply has yet been sent - that's no bad thing. Yes. But without an actual barrier it's hard to be exact - and fdatasync does the right thing by including false positives. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 12:49:35PM +0200, Wouter Verhelst wrote: > A while back, we spent quite some time defining the semantics of the > various commands in the face of the NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA > write barriers. At the time, we decided that it would be unreasonable > to expect servers to make these write barriers effective across > different connections. Do you have a nbd protocol specification? treating a flush or fua as any sort of barrier is incredibly stupid. Is it really documented that way, and if yes, why? -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 02:01:59PM +0200, Wouter Verhelst wrote: > Yes. There was some discussion on that part, and we decided that setting > the flag doesn't hurt, but the spec also clarifies that using it on READ > does nothing, semantically. > > > The problem is that there are clients in the wild which do set it on > READ, so it's just a matter of "be liberal in what you accept". Note that FUA on READ in SCSI and NVMe does have a meaning - it requires you to bypass any sort of cache on the target. I think it's an wrong defintion because it mandates implementation details that aren't observable by the initiator, but it's still the spec wording and nbd diverges from it. That's not nessecarily a bad thing, but a caveat to look out for. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Christoph, > It's not a write barrier - a write barrier was command that ensured that > > a) all previous writes were completed to the host/client > b) all previous writes were on non-volatile storage > > and > > c) the actual write with the barrier bit was on non-volatile storage Ah! the bit you are complaining about is not the bit I pointed to you, but: > NBD_CMD_FLUSH (3) > > A flush request; a write barrier. I can see that's potentially confusing as isn't meant to mean 'an old-style linux kernel block device write barrier'. I think in general terms it probably is some form of barrier, but I see no problem in deleting the words "a write barrier" from the spec text if only to make it clearer. However, I think the description of the command itself: > The server MUST NOT send a successful reply header for this request before > all write requests for which a reply has already been sent to the client have > reached permanent storage (using fsync() or similar). and the ordering section I pointed you to before, were both correct, yes? >> The point still remains that "X was sent before Y" is difficult to >> determine on the client side if X was sent over a different TCP channel >> than Y, because a packet might be dropped (requiring a retransmit) for >> X, and similar things. If blk-mq can deal with that, we're good and >> nothing further needs to be done. If not, this should be evaluated by >> someone more familiar with the internals of the kernel block subsystem >> than me. > > The important bit in all the existing protocols, and which Linux relies > on is that any write the Linux block layer got a completion for earlier > needs to be flushed out to non-volatile storage when a FLUSH command is > set. Anything that still is in flight does not matter. Which for > NBD means anything that you already replies to need to be flushed. ... that's what it says (I hope). > Or to say it more practicly - in the nbd server you simply need to > call fdatasync on the backing device or file whenever you get a FLUSH > requires, and it will do the right thing. actually fdatasync() technically does more than is necessary, as it will also flush commands that have been processed, but for which no reply has yet been sent - that's no bad thing. -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 12:52, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 12:46:07PM +0100, Alex Bligh wrote: >> Essentially NBD does supports FLUSH/FUA like this: >> >> https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt >> >> IE supports the same FLUSH/FUA primitives as other block drivers (AIUI). >> >> Link to protocol (per last email) here: >> >> https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes > > Flush as defined by the Linux block layer (and supported that way in > SCSI, ATA, NVMe) only requires to flush all already completed writes > to non-volatile media. It does not impose any ordering unlike the > nbd spec. As maintainer of the NBD spec, I'm confused as to why you think it imposes any ordering - if you think this, clearly I need to clean up the wording. Here's what it says: > The server MAY process commands out of order, and MAY reply out of order, > except that: > > • All write commands (that includes NBD_CMD_WRITE, and NBD_CMD_TRIM) > that the server completes (i.e. replies to) prior to processing to a > NBD_CMD_FLUSH MUST be written to non-volatile storage prior to replying to > that > NBD_CMD_FLUSH. This paragraph only applies if NBD_FLAG_SEND_FLUSH is set > within > the transmission flags, as otherwise NBD_CMD_FLUSH will never be sent by the > client to the server. (and another bit re FUA that isn't relevant here). Here's the Linux Kernel documentation: > The REQ_PREFLUSH flag can be OR ed into the r/w flags of a bio submitted from > the filesystem and will make sure the volatile cache of the storage device > has been flushed before the actual I/O operation is started. This explicitly > guarantees that previously completed write requests are on non-volatile > storage before the flagged bio starts. In addition the REQ_PREFLUSH flag can > be > set on an otherwise empty bio structure, which causes only an explicit cache > flush without any dependent I/O. It is recommend to use > the blkdev_issue_flush() helper for a pure cache flush. I believe that NBD treats NBD_CMD_FLUSH the same as a REQ_PREFLUSH and empty bio. If you don't read those two as compatible, I'd like to understand why not (i.e. what additional constraints one is applying that the other is not) as they are meant to be the same (save that NBD only has FLUSH as a command, i.e. the 'empty bio' version). I am happy to improve the docs to make it clearer. (sidenote: I am interested in the change from REQ_FLUSH to REQ_PREFLUSH, but in an empty bio it's not really relevant I think). > FUA as defined by the Linux block layer (and supported that way in SCSI, > ATA, NVMe) only requires the write operation the FUA bit is set on to be > on non-volatile media before completing the write operation. It does > not impose any ordering, which seems to match the nbd spec. Unlike the > NBD spec Linux does not allow FUA to be set on anything by WRITE > commands. Some other storage protocols allow a FUA bit on READ > commands or other commands that write data to the device, though. I think you mean "anything *but* WRITE commands". In NBD setting FUA on a command that does not write will do nothing, but FUA can be set on NBD_CMD_TRIM and has the expected effect. Interestingly the kernel docs are silent on which commands REQ_FUA can be set on. -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 04:38:07AM -0700, Christoph Hellwig wrote: > On Thu, Sep 15, 2016 at 12:49:35PM +0200, Wouter Verhelst wrote: > > A while back, we spent quite some time defining the semantics of the > > various commands in the face of the NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA > > write barriers. At the time, we decided that it would be unreasonable > > to expect servers to make these write barriers effective across > > different connections. > > Do you have a nbd protocol specification? https://github.com/yoe/nbd/blob/master/doc/proto.md > treating a flush or fua as any sort of barrier is incredibly stupid. Maybe I'm not using the correct terminology here. The point is that after a FLUSH, the server asserts that all write commands *for which a reply has already been sent to the client* will also have reached permanent storage. Nothing is asserted about commands for which the reply has not yet been sent, regardless of whether they were sent to the server before or after the FLUSH; they may reach permanent storage as a result of the FLUSH, or they may not, we don't say. With FUA, we only assert that the FUA-flagged command reaches permanent storage before its reply is sent, nothing else. If that's not a write barrier, then I was using the wrong terminology (and offer my apologies for the confusion). The point still remains that "X was sent before Y" is difficult to determine on the client side if X was sent over a different TCP channel than Y, because a packet might be dropped (requiring a retransmit) for X, and similar things. If blk-mq can deal with that, we're good and nothing further needs to be done. If not, this should be evaluated by someone more familiar with the internals of the kernel block subsystem than me. > Is it really documented that way, and if yes, why? The documentation does use the term write barrier, but only right before the same explanation as above. It does so because I assumed that that is what a write barrier is, and that this would make things easier to understand. If you tell me that that term is wrong as used there, I can easily remove it (it's not critical to the rest of the documentation). Regards, -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 01:55:14PM +0200, Wouter Verhelst wrote: > Maybe I'm not using the correct terminology here. The point is that > after a FLUSH, the server asserts that all write commands *for which a > reply has already been sent to the client* will also have reached > permanent storage. Nothing is asserted about commands for which the > reply has not yet been sent, regardless of whether they were sent to the > server before or after the FLUSH; they may reach permanent storage as a > result of the FLUSH, or they may not, we don't say. > > With FUA, we only assert that the FUA-flagged command reaches permanent > storage before its reply is sent, nothing else. Yes, that's the correct semantics. > If that's not a write barrier, then I was using the wrong terminology > (and offer my apologies for the confusion). It's not a write barrier - a write barrier was command that ensured that a) all previous writes were completed to the host/client b) all previous writes were on non-volatile storage and c) the actual write with the barrier bit was on non-volatile storage > The point still remains that "X was sent before Y" is difficult to > determine on the client side if X was sent over a different TCP channel > than Y, because a packet might be dropped (requiring a retransmit) for > X, and similar things. If blk-mq can deal with that, we're good and > nothing further needs to be done. If not, this should be evaluated by > someone more familiar with the internals of the kernel block subsystem > than me. The important bit in all the existing protocols, and which Linux relies on is that any write the Linux block layer got a completion for earlier needs to be flushed out to non-volatile storage when a FLUSH command is set. Anything that still is in flight does not matter. Which for NBD means anything that you already replies to need to be flushed. Or to say it more practicly - in the nbd server you simply need to call fdatasync on the backing device or file whenever you get a FLUSH requires, and it will do the right thing. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 04:52:17AM -0700, Christoph Hellwig wrote: > On Thu, Sep 15, 2016 at 12:46:07PM +0100, Alex Bligh wrote: > > Essentially NBD does supports FLUSH/FUA like this: > > > > https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt > > > > IE supports the same FLUSH/FUA primitives as other block drivers (AIUI). > > > > Link to protocol (per last email) here: > > > > https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes > > Flush as defined by the Linux block layer (and supported that way in > SCSI, ATA, NVMe) only requires to flush all already completed writes > to non-volatile media. That is precisely what FLUSH in nbd does, too. > It does not impose any ordering unlike the nbd spec. If you read the spec differently, then that's a bug in the spec. Can you clarify which part of it caused that confusion? We should fix it, then. > FUA as defined by the Linux block layer (and supported that way in SCSI, > ATA, NVMe) only requires the write operation the FUA bit is set on to be > on non-volatile media before completing the write operation. It does > not impose any ordering, which seems to match the nbd spec. Unlike the > NBD spec Linux does not allow FUA to be set on anything by WRITE > commands. Some other storage protocols allow a FUA bit on READ > commands or other commands that write data to the device, though. Yes. There was some discussion on that part, and we decided that setting the flag doesn't hurt, but the spec also clarifies that using it on READ does nothing, semantically. The problem is that there are clients in the wild which do set it on READ, so it's just a matter of "be liberal in what you accept". -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 12:46, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 12:43:35PM +0100, Alex Bligh wrote: >> Sure, it's at: >> >> https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes >> >> and that link takes you to the specific section. >> >> The treatment of FLUSH and FUA is meant to mirror exactly the >> linux block layer (or rather how the linux block layer was a few >> years ago). I even asked on LKML to verify a few points. > > Linux never expected ordering on the wire. Before 2010 we had barriers > in the kernel that provided ordering to the caller, but we never > required it from the protocol / hardware. Sure. And I think the doc section reflects exactly Linux's post-2010 expectations (note the link to the kernel documentation). IE servers are not required to order reads/writes (save that all writes that the server completes prior to reply to an NBD_CMD_FLUSH must be persisted prior to the reply to that NBD_CMD_FLUSH which I believe to be exactly the same behaviour as the kernel dealing with an empty bio with REQ_FLUSH set). Perhaps the section should be called "No ordering of messages and writes"! My point was that *in practice* disordering is not well tested as *in practice* many server implementations do in fact process each command in order, though that's changed recently. -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 12:46:07PM +0100, Alex Bligh wrote: > Essentially NBD does supports FLUSH/FUA like this: > > https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt > > IE supports the same FLUSH/FUA primitives as other block drivers (AIUI). > > Link to protocol (per last email) here: > > https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes Flush as defined by the Linux block layer (and supported that way in SCSI, ATA, NVMe) only requires to flush all already completed writes to non-volatile media. It does not impose any ordering unlike the nbd spec. FUA as defined by the Linux block layer (and supported that way in SCSI, ATA, NVMe) only requires the write operation the FUA bit is set on to be on non-volatile media before completing the write operation. It does not impose any ordering, which seems to match the nbd spec. Unlike the NBD spec Linux does not allow FUA to be set on anything by WRITE commands. Some other storage protocols allow a FUA bit on READ commands or other commands that write data to the device, though. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 12:43:35PM +0100, Alex Bligh wrote: > Sure, it's at: > > https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes > > and that link takes you to the specific section. > > The treatment of FLUSH and FUA is meant to mirror exactly the > linux block layer (or rather how the linux block layer was a few > years ago). I even asked on LKML to verify a few points. Linux never expected ordering on the wire. Before 2010 we had barriers in the kernel that provided ordering to the caller, but we never required it from the protocol / hardware. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
> On 15 Sep 2016, at 12:40, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 01:29:36PM +0200, Wouter Verhelst wrote: >> Yes, and that is why I was asking about this. If the write barriers >> are expected to be shared across connections, we have a problem. If, >> however, they are not, then it doesn't matter that the commands may be >> processed out of order. > > There is no such thing as a write barrier in the Linux kernel. We'd > much prefer protocols not to introduce any pointless synchronization > if we can avoid it. I suspect the issue is terminological. Essentially NBD does supports FLUSH/FUA like this: https://www.kernel.org/doc/Documentation/block/writeback_cache_control.txt IE supports the same FLUSH/FUA primitives as other block drivers (AIUI). Link to protocol (per last email) here: https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Christoph, > On 15 Sep 2016, at 12:38, Christoph Hellwig wrote: > > On Thu, Sep 15, 2016 at 12:49:35PM +0200, Wouter Verhelst wrote: >> A while back, we spent quite some time defining the semantics of the >> various commands in the face of the NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA >> write barriers. At the time, we decided that it would be unreasonable >> to expect servers to make these write barriers effective across >> different connections. > > Do you have a nbd protocol specification? treating a flush or fua > as any sort of barrier is incredibly stupid. Is it really documented > that way, and if yes, why? Sure, it's at: https://github.com/yoe/nbd/blob/master/doc/proto.md#ordering-of-messages-and-writes and that link takes you to the specific section. The treatment of FLUSH and FUA is meant to mirror exactly the linux block layer (or rather how the linux block layer was a few years ago). I even asked on LKML to verify a few points. -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 01:29:36PM +0200, Wouter Verhelst wrote: > Yes, and that is why I was asking about this. If the write barriers > are expected to be shared across connections, we have a problem. If, > however, they are not, then it doesn't matter that the commands may be > processed out of order. There is no such thing as a write barrier in the Linux kernel. We'd much prefer protocols not to introduce any pointless synchronization if we can avoid it. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 12:09:28PM +0100, Alex Bligh wrote: > A more general point is that with multiple queues requests > may be processed in a different order even by those servers that > currently process the requests in strict order, or in something > similar to strict order. The server is permitted by the spec > (save as mandated by NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA) to > process commands out of order anyway, but I suspect this has > to date been little tested. The Linux kernel does not assume any synchroniztion between block I/O commands. So any sort of synchronization a protocol does is complete overkill for us. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Thu, Sep 15, 2016 at 12:09:28PM +0100, Alex Bligh wrote: > Wouter, Josef, (& Eric) > > > On 15 Sep 2016, at 11:49, Wouter Verhelst wrote: > > > > Hi, > > > > On Fri, Sep 09, 2016 at 10:02:03PM +0200, Wouter Verhelst wrote: > >> I see some practical problems with this: > > [...] > > > > One more that I didn't think about earlier: > > > > A while back, we spent quite some time defining the semantics of the > > various commands in the face of the NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA > > write barriers. At the time, we decided that it would be unreasonable > > to expect servers to make these write barriers effective across > > different connections. > > Actually I wonder if there is a wider problem in that implementations > might mediate access to a device by presence of an extant TCP connection, > i.e. only permit one TCP connection to access a given block device at > once. If you think about (for instance) a forking daemon that does > writeback caching, that would be an entirely reasonable thing to do > for data consistency. Sure. They will have to live with the fact that clients connected to them will run slower; I don't think that's a problem. In addition, Josef's client implementation requires the user to explicitly ask for multiple connections. There are multiple contexts in which NBD can be used, and in some performance is more important than in others. I think that is fine. [...] > A more general point is that with multiple queues requests > may be processed in a different order even by those servers that > currently process the requests in strict order, or in something > similar to strict order. The server is permitted by the spec > (save as mandated by NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA) to > process commands out of order anyway, but I suspect this has > to date been little tested. Yes, and that is why I was asking about this. If the write barriers are expected to be shared across connections, we have a problem. If, however, they are not, then it doesn't matter that the commands may be processed out of order. [...] -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Wouter, Josef, (& Eric) > On 15 Sep 2016, at 11:49, Wouter Verhelst wrote: > > Hi, > > On Fri, Sep 09, 2016 at 10:02:03PM +0200, Wouter Verhelst wrote: >> I see some practical problems with this: > [...] > > One more that I didn't think about earlier: > > A while back, we spent quite some time defining the semantics of the > various commands in the face of the NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA > write barriers. At the time, we decided that it would be unreasonable > to expect servers to make these write barriers effective across > different connections. Actually I wonder if there is a wider problem in that implementations might mediate access to a device by presence of an extant TCP connection, i.e. only permit one TCP connection to access a given block device at once. If you think about (for instance) a forking daemon that does writeback caching, that would be an entirely reasonable thing to do for data consistency. I also wonder whether any servers that can do caching per connection will always share a consistent cache between connections. The one I'm worried about in particular here is qemu-nbd - Eric Blake CC'd. A more general point is that with multiple queues requests may be processed in a different order even by those servers that currently process the requests in strict order, or in something similar to strict order. The server is permitted by the spec (save as mandated by NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA) to process commands out of order anyway, but I suspect this has to date been little tested. Lastly I confess to lack of familiarity with the kernel side code, but how is NBD_CMD_DISCONNECT synchronised across each of the connections? Presumably you need to send it on each channel, but cannot assume the NBD connection as a whole is dead until the last tcp connection has closed? -- Alex Bligh -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Hi, On Fri, Sep 09, 2016 at 10:02:03PM +0200, Wouter Verhelst wrote: > I see some practical problems with this: [...] One more that I didn't think about earlier: A while back, we spent quite some time defining the semantics of the various commands in the face of the NBD_CMD_FLUSH and NBD_CMD_FLAG_FUA write barriers. At the time, we decided that it would be unreasonable to expect servers to make these write barriers effective across different connections. Since my knowledge of kernel internals is limited, I tried finding some documentation on this, but I guess that either it doesn't exist or I'm looking in the wrong place; therefore, am I correct in assuming that blk-mq knows about such semantics, and will handle them correctly (by either sending a write barrier to all queues, or not making assumptions about write barriers that were sent over a different queue)? If not, this may be something that needs to be taken care of. Thanks, -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 09/09/2016 05:00 PM, Josef Bacik wrote: Right. Alternatively, you could perhaps make it so that the lost connection is removed, unack'd requests on that connection are resent, and the session moves on with one less connection (unless the lost connection is the last one, in which case we die as before). That might be too much work and not worth it though. Yeah I wasn't sure if we could just randomly remove hw queue's in blk mq while the device is still up. If that is in fact easy to do then I'm in favor of trucking along with less connections than we originally had, otherwise I think it'll be too big of a pain. Also some users (Facebook in this case) would rather the whole thing fail so we can figure out what went wrong rather than suddenly going at a degraded performance. We can do that online. We do that for CPU hotplug/unplug events, and we also expose that functionality to drivers through blk_mq_update_nr_hw_queues(). So yes, should be trivial to support from nbd. -- Jens Axboe -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 09/09/2016 04:55 PM, Wouter Verhelst wrote: On Fri, Sep 09, 2016 at 04:36:07PM -0400, Josef Bacik wrote: On 09/09/2016 04:02 PM, Wouter Verhelst wrote: [...] I see some practical problems with this: - You removed the pid attribute from sysfs (unless you added it back and I didn't notice, in which case just ignore this part). This kills userspace in two ways: - systemd/udev mark an NBD device as "not active" if the sysfs pid attribute is absent. Removing that attribute causes the new nbd systemd unit to stop working. - nbd-client -check relies on this attribute too, which means that even if people don't use systemd, their init scripts will still break, and vigilant sysadmins (who check before trying to connect something) will be surprised. Ok I can add this back, I didn't see anybody using it, but again I didn't look very hard. Thank you. - What happens if userspace tries to connect an already-connected device to some other server? Currently that can't happen (you get EBUSY); with this patch, I believe it can, and data corruption would be the result (on *two* nbd devices). Additionally, with the loss of the pid attribute (as above) and the ensuing loss of the -check functionality, this might actually be a somewhat likely scenario. Once you do DO_IT then you'll get the EBUSY, so no problems. Oh, okay. I missed that part. Now if you modify the client to connect to two different servers then yes you could have data corruption, but hey if you do stupid things then bad things happen, I'm not sure we need to explicitly keep this from happening. Yeah, totally agree there. - What happens if one of the multiple connections drop but the others do not? It keeps on trucking, but the connections that break will return -EIO. That's not good, I'll fix it to tear down everything if that happens. Right. Alternatively, you could perhaps make it so that the lost connection is removed, unack'd requests on that connection are resent, and the session moves on with one less connection (unless the lost connection is the last one, in which case we die as before). That might be too much work and not worth it though. Yeah I wasn't sure if we could just randomly remove hw queue's in blk mq while the device is still up. If that is in fact easy to do then I'm in favor of trucking along with less connections than we originally had, otherwise I think it'll be too big of a pain. Also some users (Facebook in this case) would rather the whole thing fail so we can figure out what went wrong rather than suddenly going at a degraded performance. - This all has the downside that userspace now has to predict how many parallel connections will be necessary and/or useful. If the initial guess was wrong, we don't have a way to correct later on. No, it relies on the admin to specify based on their environment. Sure, but I suppose it would be nice if things could dynamically grow when needed, and/or that the admin could modify the number of connections of an already-connected device. Then again, this might also be too much work and not worth it. I mean we can set some magic number, like num_connections = min(nr_cpus, 4); and then that way people who aren't paying attention suddenly are going faster. I think anything smarter and we'd have to figure out how fast the link is and at what point we're hitting the diminishing returns and that path lies sadness. [...] A better way, long term, would presumably be to modify the protocol to allow multiplexing several requests in one NBD session. This would deal with what you're trying to fix too[1], while it would not pull in all of the above problems. [1] after all, we have to serialize all traffic anyway, just before it heads into the NIC. Yeah I considered changing the protocol to handle multiplexing different requests, but that runs into trouble since we can't guarantee that each discrete sendmsg/recvmsg is going to atomically copy our buffer in. We can accomplish this with KCM of course which is a road I went down for a little while, but then we have the issue of the actual data to send across, and KCM is limited to a certain buffer size (I don't remember what it was exactly). This limitation is fine in practice I think, but I got such good performance with multiple connections that I threw all that work away and went with this. Okay, sounds like you've given that way more thought than me, and that that's a dead end. Never mind then. Not necessarily a dead end, but a path that requires a lot of thought and experimentation to figure out if it's worth it. So maybe I'll waste an interns summer on figuring it out, but I've got other things I'd rather do ;). Thanks, Josef -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On Fri, Sep 09, 2016 at 04:36:07PM -0400, Josef Bacik wrote: > On 09/09/2016 04:02 PM, Wouter Verhelst wrote: [...] > > I see some practical problems with this: > > - You removed the pid attribute from sysfs (unless you added it back and > > I didn't notice, in which case just ignore this part). This kills > > userspace in two ways: > > - systemd/udev mark an NBD device as "not active" if the sysfs pid > > attribute is absent. Removing that attribute causes the new nbd > > systemd unit to stop working. > > - nbd-client -check relies on this attribute too, which means that > > even if people don't use systemd, their init scripts will still > > break, and vigilant sysadmins (who check before trying to connect > > something) will be surprised. > > Ok I can add this back, I didn't see anybody using it, but again I didn't > look > very hard. Thank you. > > - What happens if userspace tries to connect an already-connected device > > to some other server? Currently that can't happen (you get EBUSY); > > with this patch, I believe it can, and data corruption would be the > > result (on *two* nbd devices). Additionally, with the loss of the pid > > attribute (as above) and the ensuing loss of the -check functionality, > > this might actually be a somewhat likely scenario. > > Once you do DO_IT then you'll get the EBUSY, so no problems. Oh, okay. I missed that part. > Now if you modify the client to connect to two different servers then yes you > could have data corruption, but hey if you do stupid things then bad things > happen, I'm not sure we need to explicitly keep this from happening. Yeah, totally agree there. > > - What happens if one of the multiple connections drop but the others do > > not? > > It keeps on trucking, but the connections that break will return -EIO. > That's > not good, I'll fix it to tear down everything if that happens. Right. Alternatively, you could perhaps make it so that the lost connection is removed, unack'd requests on that connection are resent, and the session moves on with one less connection (unless the lost connection is the last one, in which case we die as before). That might be too much work and not worth it though. > > - This all has the downside that userspace now has to predict how many > > parallel connections will be necessary and/or useful. If the initial > > guess was wrong, we don't have a way to correct later on. > > No, it relies on the admin to specify based on their environment. Sure, but I suppose it would be nice if things could dynamically grow when needed, and/or that the admin could modify the number of connections of an already-connected device. Then again, this might also be too much work and not worth it. [...] > > A better way, long term, would presumably be to modify the protocol to > > allow multiplexing several requests in one NBD session. This would deal > > with what you're trying to fix too[1], while it would not pull in all of > > the above problems. > > > > [1] after all, we have to serialize all traffic anyway, just before it > > heads into the NIC. > > Yeah I considered changing the protocol to handle multiplexing different > requests, but that runs into trouble since we can't guarantee that each > discrete > sendmsg/recvmsg is going to atomically copy our buffer in. We can accomplish > this with KCM of course which is a road I went down for a little while, but > then > we have the issue of the actual data to send across, and KCM is limited to a > certain buffer size (I don't remember what it was exactly). This limitation > is > fine in practice I think, but I got such good performance with multiple > connections that I threw all that work away and went with this. Okay, sounds like you've given that way more thought than me, and that that's a dead end. Never mind then. > Thanks for the review, I'll fix up these issues you've pointed out and resend, Thanks, -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
On 09/09/2016 04:02 PM, Wouter Verhelst wrote: Hi Josef, On Thu, Sep 08, 2016 at 05:12:05PM -0400, Josef Bacik wrote: Apologies if you are getting this a second time, it appears vger ate my last submission. -- This is a patch series aimed at bringing NBD into 2016. The two big components of this series is converting nbd over to using blkmq and then allowing us to provide more than one connection for a nbd device. The NBD user space server doesn't care about how many connections it has to a particular device, so we can easily open multiple connections to the server and allow blkmq to handle multi-plexing over the different connections. I see some practical problems with this: - You removed the pid attribute from sysfs (unless you added it back and I didn't notice, in which case just ignore this part). This kills userspace in two ways: - systemd/udev mark an NBD device as "not active" if the sysfs pid attribute is absent. Removing that attribute causes the new nbd systemd unit to stop working. - nbd-client -check relies on this attribute too, which means that even if people don't use systemd, their init scripts will still break, and vigilant sysadmins (who check before trying to connect something) will be surprised. Ok I can add this back, I didn't see anybody using it, but again I didn't look very hard. - What happens if userspace tries to connect an already-connected device to some other server? Currently that can't happen (you get EBUSY); with this patch, I believe it can, and data corruption would be the result (on *two* nbd devices). Additionally, with the loss of the pid attribute (as above) and the ensuing loss of the -check functionality, this might actually be a somewhat likely scenario. Once you do DO_IT then you'll get the EBUSY, so no problems. Now if you modify the client to connect to two different servers then yes you could have data corruption, but hey if you do stupid things then bad things happen, I'm not sure we need to explicitly keep this from happening. - What happens if one of the multiple connections drop but the others do not? It keeps on trucking, but the connections that break will return -EIO. That's not good, I'll fix it to tear down everything if that happens. - This all has the downside that userspace now has to predict how many parallel connections will be necessary and/or useful. If the initial guess was wrong, we don't have a way to correct later on. No, it relies on the admin to specify based on their environment. My suggestion is to reject an additional connection unless it comes from the same userspace process as the previous connections, and to retain the pid attribute (since it is now guaranteed to be the same for all the connections). That should fix the first two issues (while unfortunately reinforcing the last one). The third would also need to have clearly defined semantics, at the very least. Yeah that sounds reasonable to me, I hadn't thought of some other pid trying to setup a device at the same time. A better way, long term, would presumably be to modify the protocol to allow multiplexing several requests in one NBD session. This would deal with what you're trying to fix too[1], while it would not pull in all of the above problems. [1] after all, we have to serialize all traffic anyway, just before it heads into the NIC. Yeah I considered changing the protocol to handle multiplexing different requests, but that runs into trouble since we can't guarantee that each discrete sendmsg/recvmsg is going to atomically copy our buffer in. We can accomplish this with KCM of course which is a road I went down for a little while, but then we have the issue of the actual data to send across, and KCM is limited to a certain buffer size (I don't remember what it was exactly). This limitation is fine in practice I think, but I got such good performance with multiple connections that I threw all that work away and went with this. Thanks for the review, I'll fix up these issues you've pointed out and resend, Josef -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [Nbd] [RESEND][PATCH 0/5] nbd improvements
Hi Josef, On Thu, Sep 08, 2016 at 05:12:05PM -0400, Josef Bacik wrote: > Apologies if you are getting this a second time, it appears vger ate my last > submission. > > -- > > This is a patch series aimed at bringing NBD into 2016. The two big > components > of this series is converting nbd over to using blkmq and then allowing us to > provide more than one connection for a nbd device. The NBD user space server > doesn't care about how many connections it has to a particular device, so we > can > easily open multiple connections to the server and allow blkmq to handle > multi-plexing over the different connections. I see some practical problems with this: - You removed the pid attribute from sysfs (unless you added it back and I didn't notice, in which case just ignore this part). This kills userspace in two ways: - systemd/udev mark an NBD device as "not active" if the sysfs pid attribute is absent. Removing that attribute causes the new nbd systemd unit to stop working. - nbd-client -check relies on this attribute too, which means that even if people don't use systemd, their init scripts will still break, and vigilant sysadmins (who check before trying to connect something) will be surprised. - What happens if userspace tries to connect an already-connected device to some other server? Currently that can't happen (you get EBUSY); with this patch, I believe it can, and data corruption would be the result (on *two* nbd devices). Additionally, with the loss of the pid attribute (as above) and the ensuing loss of the -check functionality, this might actually be a somewhat likely scenario. - What happens if one of the multiple connections drop but the others do not? - This all has the downside that userspace now has to predict how many parallel connections will be necessary and/or useful. If the initial guess was wrong, we don't have a way to correct later on. My suggestion is to reject an additional connection unless it comes from the same userspace process as the previous connections, and to retain the pid attribute (since it is now guaranteed to be the same for all the connections). That should fix the first two issues (while unfortunately reinforcing the last one). The third would also need to have clearly defined semantics, at the very least. A better way, long term, would presumably be to modify the protocol to allow multiplexing several requests in one NBD session. This would deal with what you're trying to fix too[1], while it would not pull in all of the above problems. [1] after all, we have to serialize all traffic anyway, just before it heads into the NIC. -- < ron> I mean, the main *practical* problem with C++, is there's like a dozen people in the world who think they really understand all of its rules, and pretty much all of them are just lying to themselves too. -- #debian-devel, OFTC, 2016-02-12 -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html