> So what do you think? Should we proceed with a 'hacked' version of the > message factory in 2.9 and go for the runtime message generation in later > release, or keep the code clean and fix the regression in the next releases? > Andrey, can you take a look at my change? I think it is fairly > straightforward and does not change the semantics, just skips the factory > closures for certain messages.
IMHO 2.5% isn't too much especially because it isn't actual for all workloads (I didn't get any significant drops during benchmarking). So I prefer the runtime generation in later releases. On Mon, Sep 14, 2020 at 12:41 PM Alexey Goncharuk <alexey.goncha...@gmail.com> wrote: > > Alexey, Andrey, Igniters, > > So what do you think? Should we proceed with a 'hacked' version of the > message factory in 2.9 and go for the runtime message generation in later > release, or keep the code clean and fix the regression in the next releases? > Andrey, can you take a look at my change? I think it is fairly > straightforward and does not change the semantics, just skips the factory > closures for certain messages. > > Personally, I would prefer fixing the regression given that we also > introduced tracing in this release. > > > > пт, 11 сент. 2020 г. в 12:09, Alex Plehanov <plehanov.a...@gmail.com>: >> >> Alexey, >> >> We've benchmarked by yardstick commits 130376741bf vs ed52559eb95 and the >> performance of ed52559eb95 is better for about 2.5% on average on our >> environment (about the same results we got benchmarking 65c30ec6 vs >> 0606f03d). We've made 24 runs for each commit of >> IgnitePutTxImplicitBenchmark (we got maximum drop for 2.9 on this >> benchmark), 200 seconds warmup, 300 seconds benchmark, 6 servers, 5 clients >> 50 threads each. Yardstick results for this configuration: >> Commit 130376741bf: avg TPS=164096, avg latency=9173464 ns >> Commit ed52559eb95: avg TPS=168283, avg latency=8945908 ns >> >> пт, 11 сент. 2020 г. в 09:51, Artem Budnikov <a.budnikov.ign...@gmail.com>: >>> >>> Hi Everyone, >>> >>> I posted an instruction on how to publish the docs on >>> ignite.apache.org/docs [1]. When you finish with Ignite 2.9, you can update >>> the docs by following the instruction. Unfortunately, I won't be able to >>> spend any time on this project any longer. You can send your pull requests >>> and questions about the documentation to Denis Magda. >>> >>> -Artem >>> >>> [1] : https://cwiki.apache.org/confluence/display/IGNITE/How+to+Document >>> >>> On Thu, Sep 10, 2020 at 2:45 PM Alexey Goncharuk >>> <alexey.goncha...@gmail.com> wrote: >>>> >>>> Alexey, >>>> >>>> I've tried to play with message factories locally, but unfortunately, I >>>> cannot spot the difference between old and new implementation in >>>> distributed benchmarks. I pushed an implementation of MessageFactoryImpl >>>> with the old switch statement to the ignite-2.9-revert-12568 branch >>>> (discussed this with Andrey Gura, the change should be compatible with the >>>> new metrics as we still use the register() mechanics). >>>> >>>> Can you check if this change makes any difference performance-wise in your >>>> environment? If yes, we can go with runtime code generation in the long >>>> term: register classes and generate a dynamic message factory with a switch >>>> statement once all messages are registered (not in 2.9 though, obviously). >>>> >>>> ср, 9 сент. 2020 г. в 14:53, Alex Plehanov <plehanov.a...@gmail.com>: >>>> >>>> > Hello guys, >>>> > >>>> > I've tried to optimize tracing implementation (ticket [1]), it reduced >>>> > the >>>> > drop, but not completely removed it. >>>> > Ivan Rakov, Alexander Lapin, can you please review the patch? >>>> > Ivan Artiukhov, can you please benchmark the patch [2] against 2.8.1 >>>> > release on your environment? >>>> > With this patch on our environment, it's about a 3% drop left, it's close >>>> > to measurement error and I think such a drop is not a showstopper. Guys, >>>> > WDYT? >>>> > >>>> > Also, I found that compatibility is broken for JDBC thin driver between >>>> > 2.8 >>>> > and 2.9 versions (ticket [3]). I think it's a blocker and should be >>>> > fixed in 2.9. I've prepared the patch. >>>> > Taras Ledkov, can you please review this patch? >>>> > >>>> > And one more ticket I propose to include into 2.9 [4] (NIO message >>>> > send problem in some circumstances). I will cherry-pick it if there is no >>>> > objection. >>>> > >>>> > [1] https://issues.apache.org/jira/browse/IGNITE-13411 >>>> > [2] https://github.com/apache/ignite/pull/8223 >>>> > [3] https://issues.apache.org/jira/browse/IGNITE-13414 >>>> > [4] https://issues.apache.org/jira/browse/IGNITE-13361 >>>> > >>>> > пн, 7 сент. 2020 г. в 14:14, Maxim Muzafarov <mmu...@apache.org>: >>>> > >>>> > > Alexey, >>>> > > >>>> > > I propose to include [1] issue to the 2.9 release. Since this issue is >>>> > > related to the new master key change functionality which haven't been >>>> > > released yet I think it will be safe to cherry-pick commit to the >>>> > > release branch. >>>> > > >>>> > > [1] https://issues.apache.org/jira/browse/IGNITE-13390 >>>> > > >>>> > > On Tue, 1 Sep 2020 at 12:13, Nikolay Izhikov <nizhi...@apache.org> >>>> > wrote: >>>> > > > >>>> > > > Hello, Igniters. >>>> > > > >>>> > > > Alexey, please, include one more Python thin client fix [1] into the >>>> > 2.9 >>>> > > release >>>> > > > It fixes kinda major issue - "Python client returns fields in wrong >>>> > > order since the 2 row when fields_count>10" >>>> > > > >>>> > > > [1] https://issues.apache.org/jira/browse/IGNITE-12809 >>>> > > > [2] >>>> > > >>>> > https://github.com/apache/ignite/commit/38025ee4167f05eaa2d6a2c5c2ab70c83a462cfc >>>> > > > >>>> > > > > 31 авг. 2020 г., в 19:23, Alexey Goncharuk < >>>> > alexey.goncha...@gmail.com> >>>> > > написал(а): >>>> > > > > >>>> > > > > Alexey, thanks, got it. I am not sure we can optimize anything out >>>> > > > > of >>>> > > the >>>> > > > > message factory with suppliers (at least I have no ideas right >>>> > > > > now), >>>> > so >>>> > > > > most likely the only move here is to switch back to the switch >>>> > approach >>>> > > > > somehow preserving the metrics part. Probably, inlining the Ignite >>>> > > messages >>>> > > > > to the IgniteMessageFactoryImpl should do the trick. Let me explore >>>> > the >>>> > > > > code a bit. >>>> > > > > >>>> > > > > P.S. I am surprised by the impact this part makes for the >>>> > performance. >>>> > > > > Message creation is indeed on the hot path, but a single virtual >>>> > > > > call >>>> > > > > should not make that much of a difference given the amount of other >>>> > > work >>>> > > > > happening during the message processing. >>>> > > > > >>>> > > > > пн, 31 авг. 2020 г. в 18:33, Alex Plehanov <plehanov.a...@gmail.com >>>> > >: >>>> > > > > >>>> > > > >> Alexey, sorry, I wrongly interpreted our benchmark results. >>>> > Actually, >>>> > > we >>>> > > > >> were looking for a drop using bi-sect in the range between e6a7f93 >>>> > > (first >>>> > > > >> commit in the 2.9 branch after 2.8 branch cut) and 6592dfa5 (last >>>> > > commit in >>>> > > > >> the 2.9 branch). And we found these two problematic commits. >>>> > > > >> >>>> > > > >> Perhaps only IGNITE-13060 (Tracing) is responsible for a drop >>>> > between >>>> > > > >> 2.8.1 and 2.9 (we have benchmarked 2.8.1 vs 2.9 with reverted >>>> > > IGNITE-13060 >>>> > > > >> now and performance looks the same) >>>> > > > >> >>>> > > > >> Ticket IGNITE-12568 (MessageFactory refactoring) is not related to >>>> > > drop >>>> > > > >> between 2.8.1 and 2.9, but still has some performance problem, and >>>> > we >>>> > > can >>>> > > > >> win back IGNITE-13060 drop by this ticket. >>>> > > > >> >>>> > > > >> Do we need more investigation on IGNITE-13060 or we leave it as >>>> > > > >> is? >>>> > > > >> >>>> > > > >> What should we do with IGNITE-12568 (MessageFactory refactoring)? >>>> > > > >> >>>> > > > >> пн, 31 авг. 2020 г. в 13:25, Alexey Goncharuk < >>>> > > alexey.goncha...@gmail.com >>>> > > > >>> : >>>> > > > >> >>>> > > > >>> Alexey, >>>> > > > >>> >>>> > > > >>> While investigating, I found that IGNITE-12568 has an incorrect >>>> > > > >>> fix >>>> > > > >>> version and is actually present in ignite-2.8.1 branch [1], so it >>>> > > cannot be >>>> > > > >>> the source of the drop against 2.8.1. >>>> > > > >>> >>>> > > > >>> P.S. Looks like we need to enforce a more accurate work with fix >>>> > > versions >>>> > > > >>> or develop some sort of tooling to verify the fix versions. >>>> > > > >>> >>>> > > > >>> --AG >>>> > > > >>> >>>> > > > >>> [1] >>>> > > > >>> >>>> > > >>>> > https://github.com/apache/ignite/commit/3e492bd23851856bbd0385c6a419892d0bba2a34 >>>> > > > >>> >>>> > > > >>> пн, 31 авг. 2020 г. в 12:42, Alexey Goncharuk < >>>> > > alexey.goncha...@gmail.com >>>> > > > >>>> : >>>> > > > >>> >>>> > > > >>>> пт, 28 авг. 2020 г. в 11:16, Alex Plehanov < >>>> > plehanov.a...@gmail.com >>>> > > >: >>>> > > > >>>> >>>> > > > >>>>> Guys, >>>> > > > >>>>> >>>> > > > >>>>> We have benchmarked 2.9 without IGNITE-13060 and IGNITE-12568 >>>> > > (reverted >>>> > > > >>>>> it >>>> > > > >>>>> locally) and got the same performance as on 2.8.1 >>>> > > > >>>>> >>>> > > > >>>>> IGNITE-13060 (Tracing) - some code was added to hot paths, to >>>> > trace >>>> > > > >>>>> these >>>> > > > >>>>> hot paths, it's clear why we have performance drop here. >>>> > > > >>>>> >>>> > > > >>>>> IGNITE-12568 (MessageFactory refactoring) - switch/case block >>>> > > > >>>>> was >>>> > > > >>>>> refactored to an array of message suppliers. The message >>>> > > > >>>>> factory >>>> > > is on >>>> > > > >>>>> the >>>> > > > >>>>> hot path, which explains why this commit has an impact on total >>>> > > > >>>>> performance. >>>> > > > >>>>> I've checked JIT assembly output, done some JMH >>>> > > > >>>>> microbenchmarks, >>>> > > and >>>> > > > >>>>> found >>>> > > > >>>>> that old implementation of MessageFactory.create() about 30-35% >>>> > > faster >>>> > > > >>>>> than >>>> > > > >>>>> the new one. The reason - approach with switch/case can >>>> > effectively >>>> > > > >>>>> inline >>>> > > > >>>>> message creation code, but with an array of suppliers >>>> > > > >>>>> relatively >>>> > > heavy >>>> > > > >>>>> "invokeinterface" cannot be skipped. I've tried to rewrite the >>>> > code >>>> > > > >>>>> using >>>> > > > >>>>> an abstract class for suppliers instead of an interface (to >>>> > > > >>>>> replace "invokeinterface" with the "invokevirtual"), but it >>>> > > > >>>>> gives >>>> > > back >>>> > > > >>>>> only >>>> > > > >>>>> 10% of method performance and in this case, code looks ugly >>>> > > (lambdas >>>> > > > >>>>> can't >>>> > > > >>>>> be used). Currently, I can't find any more ways to optimize the >>>> > > current >>>> > > > >>>>> approach (except return to the switch/case block). Andrey Gura, >>>> > as >>>> > > the >>>> > > > >>>>> author of IGNITE-12568, maybe you have some ideas about >>>> > > optimization? >>>> > > > >>>>> >>>> > > > >>>>> Perhaps we should revert IGNITE-12568, but there are some >>>> > > > >>>>> metrics >>>> > > > >>>>> already >>>> > > > >>>>> created, which can't be rewritten using old message factory >>>> > > > >>>>> implementation >>>> > > > >>>>> (IGNITE-12756). Guys, WDYT? >>>> > > > >>>>> >>>> > > > >>>> >>>> > > > >>>> Alexey, >>>> > > > >>>> >>>> > > > >>>> I see that IGNITE-12756 (metrics improvements) is already >>>> > > > >>>> released >>>> > > in >>>> > > > >>>> Ignite 2.8.1 while IGNITE-12568 (message factory) is only >>>> > > > >>>> present >>>> > > in Ignite >>>> > > > >>>> 2.9. Let's revert both IGNITE-12568 and whichever new metrics >>>> > > created for >>>> > > > >>>> 2.9 that depend on the new message factory to unblock the >>>> > > > >>>> release >>>> > > and deal >>>> > > > >>>> with the optimizations in 2.10? >>>> > > > >>>> >>>> > > > >>> >>>> > > > >>>> > > >>>> >
