On 06/01/14 16:33, Noel Maersk wrote:
> Around new year I started catching a sporadic segfault from
> obworker. I didn't save the backtraces (that's what drunkedness does
> :/), and can't determine what triggers it and how to reproduce (yet).
>
> If this isn't the VPS admin being naughty, or some invisible quota
> limit, then this bug made it into the release.
>
> Will get back when there's more info.
On Tue, Jan 07, 2014 at 12:09:51PM +0000, Amir Taaki wrote:
> What kind of load are you running against it? Is there some periodic
> activity that can trigger the behaviour?
OK, found it.
After running for some time (hours), obworker takes up all memory and
gets terminated.
There are sometimes streaks of these:
> ...
> ERROR [poller]: Storing block
> 000000000000001e4a1dba552d7a60d5dc649185d04521c9f5516a270f1bb7f8:
> Matching previous object found
> ...
(the block hash is different, it's not just one block)
Might be related to the same hard disk write bottleneck, but maybe not.
The server is limited to 2 GiB of RAM.
All versions are latest git, OS is Debian Wheezy, ZeroMQ version 3
(current Debian stable default).
VPS log shows this started around midnight of December 24.