Re: [APC-DEV] Re: [PECL-DEV] APC 3.1.x stability
[email protected] (shire) Wed, 21 Oct 2009 11:06:24 -0700
| Newsgroups | php.apc.dev |
|---|---|
| Message-ID | <[email protected]> |
Hi Chris/Alex/Nathan, Christopher Jones wrote: > > Forwarding to apc-dev mail list. > > Alex Michaud wrote: >> Hey all, >> I had a question about the stability of the 3.1.x branch of APC >> similar to Nathan's from this message: >> http://marc.info/?l=pecl-dev&m=124586806319746&w=2 >> I couldn't really put it better myself, and I didn't see a reply, so >> I'm copying it below. If anyone has any input on the stability issue, >> it'd be greatly appreciated! >> >> Thanks, >> Alex Michaud >> >> (quoting:) >> Apologies if this is the wrong venue to ask for support, but I >> couldn't find any other more specific APC venue. >> >> We're using APC 3.0.19 in production now and we're running into the >> deadlock issue noted on PHP bug #46025. I've done some testing to see >> if there are any configuration options we can tweak that avoid this >> deadlock, but even setting apc.write_lock = 0 still trips the >> deadlock. However, the changes made to APC 3.1.x seem to resolve the >> issue just fine (I'm assuming due to the shift to non-blocking >> semaphores rather than blocking futex locks). >> >> Obviously, I'm concerned about running "beta" software in the >> production environment, but we also can't just keep restarting Apache >> on our front-end web servers all the time either. >> >> Can anyone comment to the stability of the 3.1.x branch? Is it close >> to being tagged as stable? Are there any other specific issues with >> the new version that I should test for to see if we encounter them? FYI I just update the bug you referenced for this. If your problem is related to dead-locks due to signal handling and critical sections, it's more likely that this is a Zend issue and not so much an APC stability issue. This being said we should be releasing a new release (or RC) of APC soon, I believe what's currently in APC is fairly stable at this point. As I noted on bug #46025, if you're wiling to try the signals patch Lucas and I are working on we'd like to see if this corrects your problem so we have a verified solution. Please contact me directly if you'd like to do this and we can exchange version/patch info and any necessary debugging. -shire