Ah, you didn’t say this was not CouchDB. BigCouch does not have that fix.
Verify that you have truly raised file descriptor limit by looking at /proc/`pidof beam.smp`/limits on all nodes. 1,000’s of compactions at once is unlikely to ever work, you’ll need to manage a smaller load than that. BigCouch, like CouchDB, runs compaction at the same priority as other tasks. B. On 23 Jan 2014, at 17:21, Vladimir Ralev <[email protected]> wrote: > Thanks. I do see emfail errors all over the place indeed. As a workaround > is it feasible to enable some large system limits. I already have 1M linux > file handles limit. I am forcing compaction on a large scale - 1000s of > databases further segmented by bigcouch. I will try to pace my script next > time i run it hoping it wouldn't open all files at the same time, but still > not sure if I should touch other settings inside erlang itself. > > > On Thu, Jan 23, 2014 at 10:10 AM, Robert Samuel Newson > <[email protected]>wrote: > >> >> View indexes are not lost when compacting, but it’s possible you’re a >> release with bug COUCHDB-1445 in it ( >> https://issues.apache.org/jira/browse/COUCHDB-1445). This was fixed in >> 1.2.0 and up. >> >> B. >> >> On 22 Jan 2014, at 18:17, Vladimir Ralev <[email protected]> wrote: >> >>> Hi guys, >>> >>> I am experimenting with compaction ops. In a large DB with many big >> views a >>> compaction renders the DB unusable for almost 10 hours or so while I >> force >>> each view to be loaded by a script. Mass indexing completely overwhelms >> the >>> system. >>> >>> Is there some way to do compaction without losing the view indices? The >> way >>> I understand it indexing and compaction should be completely separate. >> >>
