SG from master + official CBL 1.0.4: WebSocket problems "libBacktraceRecording.dylib: allocate_free_list_pages() -- virtual memory exhausted!"

[Alexander Selling]

Hi! 

Ever since I upgraded to to SG from master(to get rid of the 'giving up' issue) I've started to get crashes (allocate_free_list_pages() -- virtual memory exhausted) upon full sync with the Official CBL 1.0.4. Do I have to run CBL from master?

What happens is that the web socket thread seems to be claiming memory perpetually and when it hits a couple of giga bytes the simulator crashes, I guess that'll be heavily dependent on what environment you're in.

first part of the debug window:

https://slack-files.com/T03G71GP5-F04CTHE9P-d30444d0df

second part of the debug window:

https://slack-files.com/T03G71GP5-F04C5A090-aea22e5cb9

thread backtrace all:

http://pastebin.com/cwbipvxM

[Alexander Selling]

Is there a way to force CBL to use long polling instead of web sockets? There doesn't seem to be any fallback, when I disable web sockets in nginx CBL can't sync anymore.

[Traun Leyden]

Try this:

https://github.com/couchbase/couchbase-lite-ios/issues/246#issuecomment-57250943

Sent from my iPhone

--
You received this message because you are subscribed to the Google Groups "Couchbase Mobile" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mobile-couchba...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mobile-couchbase/6c4fd92a-e1e6-47b4-8a1b-3462981b77b4%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Alexander Selling]

Awesome!

Any chance that you also know how to slow down the poll frequency? it's shooting away a million requests to _changes right now, keeping my cpu at a steady 70% load.

[Jens Alfke]

On Apr 13, 2015, at 5:07 AM, Alexander Selling <alexande...@gmail.com> wrote:

Ever since I upgraded to to SG from master(to get rid of the 'giving up' issue) I've started to get crashes (allocate_free_list_pages() -- virtual memory exhausted) upon full sync with the Official CBL 1.0.4. Do I have to run CBL from master?

Please always indicate what platform you’re using. (In this case I figured out it’s iOS.) We have three client implementations now and I only contribute to one of them, so it gets hard to triage the incoming support requests when I can’t figure out whether a message applies to me or not.

What happens is that the web socket thread seems to be claiming memory perpetually and when it hits a couple of giga bytes the simulator crashes, I guess that'll be heavily dependent on what environment you're in.

Please also report bugs using the Github issue tracker. It’s fine to also bring them up on the list, posting a link to the issue. But problems that only get reported on the mailing list can fall between the cracks and not get tracked properly for sprint planning, and the continued discussion wastes bandwidth for list members who don’t find the issue relevant.

Once you’ve filed an issue I’ll be glad to look at it.

—Jens

[Adam Fraser]

Hi Alexander,

In either case (websockets or longpolling), I'd like to better understand the Sync Gateway behaviour you're seeing.  Both cases sound a bit like SG is repeatedly sending the same changes results (resulting in excessive traffic on the websocket, and high request iteration for longpoll).  If that's the case, I second Jens' suggestion that you file a github ticket with some additional details so that we can try to reproduce, and fix if it's an issue on the Sync Gateway side.

Thanks!

Adam