*** Osso has quit IRC | 00:55 | |
*** robert_ has joined #circuits | 02:04 | |
*** robert_ has quit IRC | 02:04 | |
*** robert_ has joined #circuits | 02:04 | |
y0no | prologic, issue opened: https://github.com/circuits/circuits/issues/54 | 07:29 |
---|---|---|
*** treemo has joined #circuits | 07:40 | |
koobs | prologic: /join #freebsd-docker in case you missed it on twitter | 07:49 |
koobs | prologic: spose you couldnt really miss it with all the twitter notifications youre getting though ;) | 07:51 |
prologic | y0no, I saw :) thank you | 08:17 |
prologic | Q: Is there a way we can take into account the test's own memory usage | 08:17 |
prologic | and take that out of the equiation/graphing? | 08:17 |
y0no | prologic, we can read memory one time before call and just after to get only callEvent consumption | 08:21 |
y0no | but it will be more compilcated to get fireEvent consumption with this | 08:21 |
prologic | hmm | 08:22 |
prologic | let me think about it | 08:22 |
y0no | I think that the effect of list usage inside the test case is inconsiderable. | 08:22 |
prologic | I think it's possible | 08:22 |
prologic | subclass Manager | 08:22 |
prologic | and override run() | 08:22 |
prologic | wrap it in a decorator that captures the memory used before and after each tick | 08:23 |
prologic | but I love what you guys have done | 08:23 |
y0no | Another solution can be to measure call + list, and list only and do the difference between results. | 08:23 |
prologic | but I'd really like to see it in a test-case kind of form | 08:23 |
prologic | so we never leak memroy again | 08:23 |
prologic | and to take into account the test's own memory | 08:23 |
prologic | yeah | 08:23 |
prologic | anyway | 08:24 |
prologic | do we know where the memory leaks are coming from? | 08:24 |
prologic | you mentioned our coro stuff is leaking memory | 08:24 |
prologic | which is kind of new and not surprising | 08:24 |
prologic | but I'm surprised there's a memory leak anywhere else tbh | 08:24 |
prologic | unless it was accidently introduced in 2.2+ | 08:24 |
prologic | I have many of my websites running in 2.1 still :( | 08:24 |
prologic | and they're backends have been up for a long time :) | 08:24 |
prologic | bbs | 08:25 |
y0no | In our production system we use a lot of tasks, a lot of events, etc... And we have seen memory issue after 200,000 events processed. | 08:26 |
y0no | It seem that the finished event are not released from memory. SO if you use little processing on event you will not be impacted, but if the processing is important you will see your memory explode after few hours. | 08:27 |
y0no | As you can see on my second comments on the issue, we have found with treemo a first improve of memory usage. But this is currently not the best solution because we still have memory leak on our production, lesser than before but still there. | 08:29 |
prologic | ahh | 08:46 |
prologic | good to see real-world stuff there | 08:46 |
prologic | impressive too I might add | 08:46 |
prologic | nice :) | 08:46 |
prologic | so here's the thing | 08:46 |
prologic | let's fix this ASAP :) | 08:47 |
prologic | and see (if it's not too much effort) to ensure we don't introduce these kinds of memory leaks again | 08:47 |
y0no | prologic, yep it seem good | 09:45 |
y0no | the other priority is to fix the tests | 09:46 |
y0no | currently the build is not ok :/ | 09:46 |
*** Osso has joined #circuits | 10:23 | |
prologic | yeah | 10:41 |
prologic | but it's only one or two failing tests | 10:41 |
prologic | we're okay :) | 10:41 |
prologic | we'll get it green real soon :) | 10:41 |
prologic | and hopefully cut a release soon too | 10:41 |
prologic | we need help of course so keep those contributiosn coming! | 10:41 |
prologic | time and man power et all :) | 10:41 |
*** Coldblackice has quit IRC | 13:25 | |
*** treemo has quit IRC | 17:17 | |
*** treemo has joined #circuits | 17:43 | |
*** treemo has quit IRC | 17:43 | |
*** kwmiebach has quit IRC | 23:21 | |
*** kwmiebach has joined #circuits | 23:27 | |
*** Osso has quit IRC | 23:35 |
Generated by irclog2html.py 2.11.0 by Marius Gedminas - find it at mg.pov.lt!