If this is not a known problem I will file a new issue.
I've got two things that are controlled by once-a-minute schedules -
updating my Solr indexes, and refreshes on a servlet status page that
aggregates stats from many Solr cores into a compact table format. Both
use SolrJ and run on Linux. Both are scheduled to happen at the top of
the minute.
There is typically some very small discrepancy between when these two
processes kick off - the update program is a standalone java application
that uses the time on a Linux server with NTP, the servlet is refreshed
by javascript, usually running on Firefox for Windows, subject to
Microsoft's lousy time synchronization. At any given time, there are
usually one or two browsers out there with this page open.
The problem I'm having: Sometimes the request to /admin/mbeans happens
while an update/commit is happening, and that is resulting in exceptions
on branch_4x revision 1421496. I have seen this problem before, but it
was happening at the same time I was experiencing SOLR-4135, so I
couldn't be sure it was not a result of that issue. Now SOLR-4135 is
resolved, and I don't see THOSE exceptions any more. Below is a URL
where you can get the log showing the exceptions I am seeing now. When
the problem happens, there are four exceptions in quick succession. The
first two of those four appear to be identical.
https://dl.dropbox.com/u/97770508/solr-2012-12-14.log
The servlet is talking to four 3.5.0 servers in addition to my dev
branch_4x, and this problem is NOT happening on the 3.5.0 servers.
Thanks,
Shawn
p.s. Additional info not relevant to the issue, but might interest some:
The servlet is very fast - it gathers data in parallel from
/admin/mbeans on 47 cores and /admin/ping on 5 cores (spread across 5
servers) and renders HTML using that data, all in under half a second.
You may look at those numbers and express concern that if hundreds of
browsers have the page open, it will overload Solr with requests. I've
taken steps to ensure that doesn't happen. The data gathering portion
of the code is synchronized on a servlet-wide static object, and if the
last successful data gathering was less than five seconds ago, it will
skip the process and use existing data.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
