Vladimir Smirnov
Vladimir Smirnov
> I handle all libevent stuff in only one thread, and i have validated it with helgrind and drd, so i don't expect any thread-safety issues. Yeah, I only meant...
I just wanted to say that we ended up rebasing @chbuescher patchset on top of v2.6 (the last one that was stable for us) and it helped to solve one...
PCRE was just nice to have, that's it. It haven't helped our case a lot. Though I think while porting that to 2.6 I've accidentally reverted some fixes for 'send...
In the version that @chbuescher patched. For some reason it resurrects one of the old bugs when statistics is malformed.
@chbuescher well, we haven't saw this bug in our production environment. I've tried to run your fork, it segfaults on shutdown with the following backtrace: ``` ================================================================= ==89509==ERROR: AddressSanitizer: SEGV...
That bug with malformed sends doesn't happen if I use ```memchr``` instead of ```memrchr```. EDIT: no it haven't, just delayed the time before problem appeared again
The way how the conn->buff fills allows it to have incomplete metrics at this point, but only in case if the batch is large. I'm able to reproduce that by...
I think it's writing data to that pipe in one thread, but there are some problems with how you store partial metrics.
Yeah, sending over TCP seems to fix the problem
If I'm correct graphite-web 0.9.x and graphite-web@master have different format to answer those questions. There is an effort to make carbonserver compatibility layer (basically carbonserver with some fixes, cache support...