[Coke] Where do you see the runs? 00:20
I see that pipelines is available in Raku org, but zero evidence of any runs in raku/nqp 00:29
ah. randomly guessed dev.azure.com/raku 00:34
ok; I removed the pipeline app from the org. Now I have no idea how to re-add it. :| 00:40
ok, in azdo, add a new pipeline. But it wants to do it as me, with access to all the orgs I have access to (not just raku) 00:43
dev.azure.com/Raku/nqp/_build/resu...ew=results 00:45
new job queued 00:46
I setup *only* raku/nqp
00:56 JimmyZhuo joined
JimmyZhuo timo: thanks for review 00:57
01:02 librasteve_ left
[Coke] job mostly done, mac jvm failed, win/lin jvm still running, everything else green 01:08
patrickb: let me know if it's cool?
01:10 JimmyZhuo left 05:03 [Coke] left, [Coke] joined
patrickb Coke: I believe it is finally done! Thank you so much, 07:00
ab5tract some time ago I was a bit harsh about NQP::Config in this channel 08:25
I would like to retract my statements and apologize
It's actually a good example of a software that Just Works and doesn't need much in the way of maintenance
09:14 bisectable6 left 09:15 evalable6 left 09:16 evalable6 joined, unicodable6 left, unicodable6 joined 10:05 melezhik_ joined
melezhik_ o/ 10:05
I've added whateverabke support to brownie, now one can test against specific rakudo commit - brw.sparrowhub.io/project/brw-orch 10:06
rakudo_version input field should be 40 symbols git commit sha 10:07
login: admin password: admin 10:08
cc ab5tract: [Coke]:
10:13 melezhik_ left, melezhik_ joined 10:18 melezhik_ left, melezhik_ joined 10:21 melezhik_ left
lizmat m: dd True ?| (say "foo") # /me wonders whether infix ?& ?| ?^ should short-circuit 11:44
camelia foo
Bool::True
11:47 melezhik_ joined
melezhik_ run zef install Kind against latest Rakudo commit 11:47
brw.sparrowhub.io/report/brw-orch/120
11:52 melezhik_ left 12:27 melezhik_ joined
melezhik_ [Coke]: currently there is just a one agent in pool ( which runs from my laptop and so often is offline ) 12:28
run it 7 threads though ), so will try to run another 100 random modules and get timing ... 12:31
12:44 melezhik_ left, melezhik_ joined 12:48 melezhik_ left 12:55 melezhik_ joined
lizmat m: dd :[ 1,2,3 ] # wonder whether this should be a compilation error, it's really invalid syntax at this moment 12:55
camelia [1, 2, 3] 12:56
13:00 melezhik_ left 13:02 melezhik_ joined 13:06 melezhik_ left 13:11 melezhik_ joined 13:21 melezhik_ left, melezhik_ joined 13:32 [Tux] left 13:34 lizmat_ joined 13:36 [Tux] joined 13:37 lizmat left 13:39 melezhik_ left
releasable6 Next release in ā‰ˆ4 days and ā‰ˆ3 hours. There are no known blockers. Please log your changes in the ChangeLog: github.com/rakudo/rakudo/wiki/ChangeLog-Draft 15:00
[Coke] ā³ 343 out of 2385 modules processed (14.38%) 15:16
ā³ 533 out of 2385 modules processed (22.34%) 16:26
16:30 coleman left 16:32 coleman joined
[Coke] crash 17:04
very annoying that blin not only crashes but *kills the shell that was running it*
had another tmux window open this time, and while it stayed open, the original window just closed out. 17:05
timo [Coke]: if it's oom-related, you do have some control over how process trees get killed, via cgroups 17:06
[Coke] I see "Running Ad-Hoc Processes in cgroups" on documentation.ubuntu.com/security/...n/cgroups/ - this is showing me how to set limits but I'm not seeing control over how they get killed, just why 17:12
er, when 17:13
timo right, systemd-run is a convenient tool to get a fresh cgroup for a command
i think tmux might have cgroups set up for processes in its panes
does something relevant show up in `systemd-cgls`?
i see a cgroup called "tmux-spawn-[uuid].scope inside my app.slice 17:14
[Coke] ... also it makes me enter a password when running that. I don't think I *have* the password for this account, just an ssh key 17:15
timo that might depend on ... many things 17:16
[Coke] "too magic for my blood"
timo try `systemd-run --user ...` to communicate with your user's very own systemd instance, instead of the system-wide one
[Coke] Running as unit: blin.service; invocation ID: c674dd29d729405bbbbd90eabf6bdba2 17:18
... and it immediately returned, no output from the process...
timo haha, right
it's a bit like "docker run", if you don't ask for it to hook up to your tty and be interactive, it won't
you'll see what it does in the journal, but you probably want it directly in your terminal? 17:19
[Coke] "the journal"?
timo the --property / -p flag lets you add properties just like the `systemctl set-property` command
yeah `journalctl -f` to follow it 17:20
[Coke] Nov 11 17:17:52 raku-blin raku[105144]: Could not open bin/blin.p6. Failed to stat file: no such file or directory
timo ah, it also doesn't inherit your environment by default i assume? 17:21
[Coke] I have no idea - you're the one teaching me right now. :)
timo are you passing -d aka --same-dir? that could be the issue actually 17:22
[Coke] Best case, it looks like this will let me kill blin, leaving the tmux window open - that still doesn't really help
RAKULIB=. systemd-run --user --unit=blin --slice=user.slice --property="CPUWeight=100" --property="MemoryMax=12G" raku - 17:23
-ll-exception bin/blin.p6 --old=$OLD --new=$NEW --nproc-multiplier=$NPROCMULT --heartbeat=$HEARTBEAT 2>&1 | tee blin.out
I will try throwing more hardware at this, but that'll be later. 17:24
(the tee there is useless, will remove that)
This seems like a lot of setup to try something that doesn't feel like it's going to add any value. 17:25
(and it takes over an hour for the blin run to fail, so I don't even get quick turnaround here.) 17:26
timo to make sure the RAKULIB=. makes it in, put it as -E RAKULIB=. as an argument to systemd-run
[Coke] Will just throw more hardware at it this week and hope that I don't have to do another release before brownie is done. 17:27
timo and to get the output in your terminal and such, pass --tty --wait
[Coke] Failed to start transient service unit: Unit blin.service was already loaded or has a fragment file. 17:29
timo ah, haha, i guess that's what you get if you don't --collect :D
i guess it has to be manually unloaded? never had to do that myself i don't think, because by default it gives the unit a random name 17:30
maybe systemctl reset-failed blin.service does it? 17:31
[Coke] "systemctl --user reset-failed" worked to clear it
timo oh, right, --user will also always have to be passed to systemctl to interact with the@s
[Coke] ok, added working-directory, it's now running 17:33
Now it'll take 5m for it to get to a point where I can see if I've over-limited the cpu
timo you can find all the properties that the thing currently has, for inspiration what you may be interested in changing, with `systemctl --user show --all blin.service`
[Coke] ... thats over 400 lines long 17:35
timo yup!
[Coke] after one screen, my brain went smooth
timo it's only 200 for my tmux-spawn slice
OOMPolicy= could be interesting 17:36
[Coke] OOMPolicy=stop 17:37
17:37 melezhik_ joined
[Coke] askubuntu suggesting maybe "sudo apt remove systemd-oomd 17:38
timo if i understand right, stop means that if anything inside the unit is oom-killed, systemd will stop the unit for you. another option is OOMPolicy=continue which lets the unit continue running
removing oomd takes away some control over how OOM happens, the kernel will still oom-kill stuff for you happily 17:39
[Coke] oh. Mem: 15942 6650 8666 8 979 9291
timo though note that the control it gives includes making kills happen earlier
[Coke] Swap: 4095 494 3601
timo that looks like 9 gigs of ram available? 17:40
[Coke] 16G total
(blin is currently running, so we're using about 6 as we start up)
but I'm wondering if I cheaped out on the swap
timo do you have zram swap as well as a swap file or partition? 17:41
[Coke] ?
timo if you have `zramctl`, it'll output if there's any zram thingies, on my system it has one with "MOUNTPOINT" "[SWAP]" 17:42
[Coke] zramctl -> says nothing
timo i have 6.5 gigabytes worth of data in mine, but they only take up 3.3 gigs thanks to its compression 17:43
[Coke] afk for a bit 17:44
17:45 melezhik2 joined
timo there's a lot of neat stuff you can do if you harness the features systemd offers 17:45
i was thinking that rakudo could by default log to the journal when it's doing precomp work, for example
melezhik2 . 17:46
[Coke]: are you interested in testing out brownie ?
17:47 melezhik2 left
melezhik_ . 17:47
17:48 melezhik joined
timo maybe not immediately before a release? 17:48
melezhik_ I mean we can do it in parallel and not necessarily on all modules 17:49
which commit are being tested right now on blin, btw? 17:51
timo i assume it'd be main and the NQP_REVISION and MOAR_REVISION
melezhik_ ok 17:53
timo that's why release managers tend to become a little bit annoyed when someone merges to rakudo main very shortly before a release while blin is blinning 17:56
17:59 melezhik_ left 18:24 melezhik_ joined 18:30 melezhik_ left
Geth rakudo/andthen-orelse: 10 commits pushed by ab5tract++
review: github.com/rakudo/rakudo/compare/1...23bc1c2703		 18:32
ab5tract melezhik: I'm getting ready to try a run on the commit from e523bc1c2703c63d1639b5095f65afca85c38f20
sorry, from R#5917 18:33
linkable6 R#5917 [open]: github.com/rakudo/rakudo/pull/5917 New andthen/orelse for 6.e
18:33 melezhik_ joined
melezhik_ . 18:33
ab5tract melezhik_: o/ 18:34
18:34 lizmat_ left
melezhik_ ab5tract: hi! 18:34
ab5tract I'd like to test e523bc1c2703c63d1639b5095f65afca85c38f20 -- exactly because I want to know how rude it would be to merge right before a release ;)
18:34 lizmat joined
melezhik_ I thought something happened on the way to your trip )) 18:34
ok, running o10r on it 18:35
100 modules
brw.sparrowhub.io/report/brw-orch/2
please make sure agent build from the latest commiy 18:36
commit
ab5tract alright. locally I have 1 agent with 7 jobs / 7 cores / 12612 mb
melezhik_ cool
100 modules is enough ?
run mine agent on my laptop with 8 threads as well 18:38
ok - download whateverable.6lang.org/e523bc1c27 ... FAILED 18:40
is it already on whateverable ? 18:41
ah, this PR commit , not published to main ...
sigh
ab5tract ah, I thought it tracked off-main commits
it might be due to my just pushing it very recently? 18:42
melezhik_ in that case it needs to compile them
ab5tract worth trying again in a few minutes I think
melezhik_ I am not sure if PR's commits even are punished to whateverable?
published 18:43
[Coke] Finished with result: oom-kill 18:50
c: e523bc1c27 3.put 18:51
committable6 [Coke], Ā¦e523bc1: Ā«Cannot find this revision (did you mean ā€œ5733721ā€?)Ā»
lizmat [Coke]: as soon as you have an idea which module is OOMing, lemme know
[Coke] ^^ that's how you can test here if a revision has a build
guessing it's one of these, lizmat : gist.github.com/coke/34eb7fc14ca77...02a0073fb7 18:53
as I am only testing 8 simultaneously. 18:54
melezhik_ ab5tract: worth to add compilation for commit into brownie?
lizmat Serialize-Naive fails for me because of lack of a META6.json file? 18:59
[Coke] could be, that listing is just showing "about to test", not "tested" 19:00
lizmat Slang:Date is also faulty (only a single :)
[Coke] we have 1 commit in moarvm that needs to be pulled into nqp/rakudo
(3 altogether, but one that changes code) 19:01
lizmat those are the ones that fail for me, but nowhere near an OOM
(on HEAD)
[Coke] bah
lizmat++ for checking
melezhik_ it could be memory leak on blin overall maybe ? just a thought ... 19:02
c: e523bc1c27 3.put 19:07
committable6 melezhik_, Ā¦e523bc1: Ā«Cannot find this revision (did you mean ā€œ5733721ā€?)Ā»
lizmat if it's a leak in blin, perhaps try with a different set of modules ?
19:19 melezhik_ left, melezhik_ joined
lizmat [Coke]: one thing I just realized, with 64G RAM I guess it won't OOM for me any time soon 19:21
19:25 melezhik_ left, melezhik_ joined
[Coke] melezhik_: it's only started occuring the last 2 releases. 19:26
I will probably build a new azure vm with more memory tonight. :| 19:27
(no need to jump through hoops if we're trying to switch away)
19:30 melezhik_ left, melezhik_ joined
[Coke] for some reason I thought I had maxed out something here, but I can just pick a bigger box and press "resize" 19:34
going for 8 core, 64G
(was at 8/16)
.. wow, took about 2 seconds and a restart to get it back 19:36
yay cloud
20:07 melezhik_ left 20:38 melezhik left
lizmat [Coke]: would you be ok with a RakuAST commit at this stage? 20:44
[Coke] as long as it passes tests and you also bump the moarvm/nqp revs. :) 20:45
If this run works, I'll do another run tomorrow JIC.
going afk for a few hours, ta
lizmat it wouldn't need a dump in MoarVM or NQP 20:46
pure nqp in the ast dir
ab5tract melezhik_: I think it would be worth it. Usually when I want to run a blin, it's before a PR merge 20:47
tellable6 ab5tract, I'll pass your message to melezhik_
ab5tract not necessarily the normal use case for it, I guess, but it's usually the moment when I want to test the impact of changes 20:48
22:04 melezhik_ joined 22:06 melezhik_ left 23:05 melezhik_ joined 23:07 melezhik_ left
[Coke] aaaand it ran out of memory again - and of course I didn't run it with timo's stuff so it just died at some point while I was afk. 23:31
rerunning with the system thing.
[Coke] wonders if it would kill the shell if he ran it OUTSIDE of tmux, or if that's a "feature" of the tmux setup 23:42
timo [Coke]: also check the tmux option "remain-on-exit" 23:47
23:54 melezhik_ joined 23:57 melezhik_ left

Powered by Cro and the RakuĀ® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!