I've finally managed to reproduce this problem in a test environment, and
filed a bug report:
https://issues.apache.org/jira/browse/COUCHDB-1100
Best regards,
Pasi
On Sat, Feb 26, 2011 at 11:19, Pasi Eronen p...@iki.fi wrote:
BTW, it seems just changing ulimit -n did not work. At some point,
BTW, it seems just changing ulimit -n did not work. At some point,
the server just stopped responding completely (even the root URL just
times out, and there's nothing in the error log).
A quick googling found references to enabling kernel poll (at build time
and on erl command line), and
Hi,
I had a big batch job (inserting 10M+ documents and generating views for them)
that ran just fine for about 6 hours, but then I got this error:
[Thu, 24 Feb 2011 19:42:57 GMT] [error] [0.276.0] ** Generic server
0.276.0 terminating
** Last message in was delayed_commit
** When Server state
On Fri, Feb 25, 2011 at 4:18 AM, Pasi Eronen p...@iki.fi wrote:
Hi,
I had a big batch job (inserting 10M+ documents and generating views for them)
that ran just fine for about 6 hours, but then I got this error:
[Thu, 24 Feb 2011 19:42:57 GMT] [error] [0.276.0] ** Generic server
0.276.0
Thanks! I looked at /proc/(pid-of-beam.smp)/fd, and it seems CouchDB
is indeed using a lot of file descriptors, and the number is growing.
I restarted CouchDB (and my batch job), and after 40 minutes, it was
already using 300 fds; an hour later, the figure was 600; half
an hour later, almost 700.
It sounds like your client is doing something funky with shutting down
connections to the server. If I were to guess I would say that it
looks like you're leaking sockets between a call to shutdown and a
call to close. Though to say for certain I'd need to see what state
both ends of the socket
Yes, the client is doing something funky :-) Many of the requests
it sends are POSTs with chunked transfer encoding, so it's being
hit hard by issue COUCHDB-682. In other words, there's lot of closing
and opening of TCP connections going on, and retrying requests that
failed due to COUCHDB-682.
On Feb 25, 2011, at 7:31 PM, Pasi Eronen p...@iki.fi wrote:
Yes, the client is doing something funky :-) Many of the requests
it sends are POSTs with chunked transfer encoding, so it's being
hit hard by issue COUCHDB-682. In other words, there's lot of closing
and opening of TCP