2021-03-06 10:25 CET

View Issue Details Jump to Notes ]
IDProjectCategoryView StatusLast Update
0005992Spring engineGeneralpublic2018-09-05 20:13
Assigned To 
StatusclosedResolutionno change required 
Product Version104.0 +git 
Target VersionFixed in Version 
Summary0005992: 104.0.1-527-gf2536df maintenance Some sort of performance regression
DescriptionPerformance seems worse since I last tested. See:
 * https://www.youtube.com/watch?v=nxfnZD1wmFA
 * https://www.youtube.com/watch?v=VkHPGDbH_Yg
TagsNo tags attached.
Checked infolog.txt for Errors
Attached Files




Kloot (developer)

Last edited: 2018-05-30 20:44

View 3 revisions

the sim grapher chunks are much larger despite no significant changes so the cause won't be anything conventional, bisect.


Google_Frog (reporter)

KingRaptor has bisected and supplied some debug view screenshots https://drive.google.com/file/d/14HffKsSJ8ltpxho2ANFc70kIRzDKj7_H/view

Their feedback is the suspicion that the issue is between 287 and 309 or 525 and 527.


Kloot (developer)

525-527 are exceedingly unlikely to explain this.

309-a8b5ffc is the Lua gc change from 30/s to 1/f we already discussed in 0005951, still in effect as of 547-g876a1a1 but with randomized dropout since 477-g706512b.


Google_Frog (reporter)

How about reverting the Lua GC changes to test whether they are the cause? I see only good outcomes:
 - If the regression persists then we know to look elsewhere.
 - If the regression goes away the source is certain and the rest of the engine can be tested.


silentwings (reporter)

Wouldn't that result in the same crashes on resuming that appeared previously?


Google_Frog (reporter)

"crash on resuming"? Do you mean catchup OOM? The performance regression is much worse than the OOM crashes, as show by the lack of complaining from players about OOM compared to the significant complaints about performance when on 104-0000298:0000358. We don't know if dev even fixes the OOM.


silentwings (reporter)

No wish to stop you testing, but my estimation is that OOM caused way more issues in BA than any perf issue. (Also, I am not even convinced that a perf issue exists for BA - I record fps/gamespeed data from all players whenever I'm online.)


silentwings (reporter)

Oh also, "crash" because the OOM crashes luarules in BA, and causes desyncs, which seriously hurts players confidence in engine, if/when it happpens.


Google_Frog (reporter)

Where is this estimation from? As far as I know BA is on 104.0.1-151 and that (I heard a rumor that) there are dev hosts that don't see much use. Tell me a string (of letters or numbers) which only occurs in OOM logs and I'll be able to count the instances of OOM on 104-350ish.

I think your comments are not particularly relevant because the aim would be to find out if the GC change is responsible for the performance regression. You have not provided any hard evidence that this test not worth doing.

I am also having some trouble reading your comments, perhaps due to typos. I don't know what "No wish to stop you testing" is meant to mean and with "your" the sentence does not make much sense either.


Google_Frog (reporter)

Here is the rumor, for reference. From #zkdev:
"Ruwetuin - Yesterday at 04:56
ba is runnign the latest engine, its just not played by the players because there are always some issues, and drama about them
but its heading in the right direction i think"


Google_Frog (reporter)

Zero-K was on a version with the GC change for about five weeks. Most of that was on 104.0.1-358-gb58f0bb. I do not see much change in the rate of OOM errors. Here are the crash reports from that time filtered for spring_lua_alloc_log_error: https://github.com/ZeroK-RTS/CrashReports/issues?page=7&q=is%3Aissue+is%3Aopen+spring_lua_alloc_log_error&utf8=%E2%9C%93


Kloot (developer)

Last edited: 2018-06-09 17:17

View 2 revisions

You can use /luagccontrol in 556-g20c5c96 to switch between 30/s and 1/f modes at runtime.

Since the bizarro universe you live in apparently values constant crashes and desyncs over an unsubstantiated performance loss with no working theory behind it, and because I don't imagine much thought was put into the decision to roll back months worth of development time and effort based solely on rumors (let alone consideration of how such a move might be perceived here) or see an end to the constant second-guessing of misunderstood engine changes, I am no longer willing to do anything else on ZK's behalf from now on. Have fun making the same mistakes as during the 91 era all over again.


silentwings (reporter)

> Where is this estimation from?
I record fps/gamespeed data from all players whenever I'm online.

> You have not provided any hard evidence that this test not worth doing.
> "No wish to stop you testing"
-> I have no wish to stop you testing.

> Tell me a string (of letters or numbers) which only occurs in OOM logs and I'll be able to count the instances of OOM on 104-350ish.
I have no infologs of this and, since BA has moved on, am unlikely to ever have any.

-Issue History
Date Modified Username Field Change
2018-05-28 11:41 Google_Frog New Issue
2018-05-28 11:41 Google_Frog File Added: infolog.txt
2018-05-28 11:41 Google_Frog File Added: 20180528_192854_Comet Catcher Redux v3_104.0.1-287-gf7b0fcc maintenance.sdfz
2018-05-28 11:41 Google_Frog File Added: 20180528_190219_Comet Catcher Redux v3_104.0.1-527-gf2536df maintenance.sdfz
2018-05-30 20:36 Kloot Note Added: 0019144
2018-05-30 20:43 Kloot Note Edited: 0019144 View Revisions
2018-05-30 20:44 Kloot Note Edited: 0019144 View Revisions
2018-06-03 06:06 Google_Frog Note Added: 0019150
2018-06-03 22:15 Kloot Note Added: 0019160
2018-06-08 05:12 Google_Frog Note Added: 0019197
2018-06-08 17:23 silentwings Note Added: 0019199
2018-06-09 04:53 Google_Frog Note Added: 0019200
2018-06-09 10:08 silentwings Note Added: 0019201
2018-06-09 10:10 silentwings Note Added: 0019202
2018-06-09 10:35 Google_Frog Note Added: 0019203
2018-06-09 10:36 Google_Frog Note Added: 0019204
2018-06-09 11:17 Google_Frog Note Added: 0019205
2018-06-09 17:12 Kloot Note Added: 0019206
2018-06-09 17:17 Kloot Note Edited: 0019206 View Revisions
2018-06-09 17:19 silentwings Note Added: 0019207
2018-09-05 20:13 Kloot Status new => closed
2018-09-05 20:13 Kloot Resolution open => no change required
+Issue History