2009-07-13
§
|
21:31 |
<Rob> |
pushing dns update to fix management ips for new apaches |
[production] |
19:05 |
<Fred> |
added storage3 to ganglia monitor. |
[production] |
18:50 |
<brion> |
synchronized php-1.5/abusefilter.php 'Disable dewiki missingsummary, mysteriously in abusefilter section. Per bug 19208' |
[production] |
16:30 |
<Fred> |
install wikimedia-nis-client to srv66 and mounted /home. |
[production] |
16:28 |
<brion> |
synchronized php-1.5/InitialiseSettings.php 'fixing wikispecies RC-IRC prefix to species.wikimedia' |
[production] |
16:27 |
<brion> |
test wiki was apparently moved from dead srv35 to srv66, which has new NFS-less config. thus fail since test runs from nfs |
[production] |
16:24 |
<brion> |
test wiki borked; reported down for several days now :) investigating |
[production] |
15:12 |
<midom> |
synchronized php-1.5/db.php 'db26 raid issues' |
[production] |
14:55 |
<midom> |
synchronized php-1.5/db.php 'db3 and db5 coming live as commons servers' |
[production] |
14:13 |
<domas> |
dropped few more snapshots, as %sys was increasing on ms1... |
[production] |
11:16 |
<domas> |
manually restarted plethora of failing apaches (direct segfaults and other possible APC corruptions, leading to php OOM errors) |
[production] |
09:50 |
<tstarling> |
synchronized php-1.5/includes/specials/SpecialBlockip.php |
[production] |
09:00 |
<Tim> |
restarted apache2 on image scalers |
[production] |
08:39 |
<tstarling> |
synchronized php-1.5/includes/Math.php 'statless render hack' |
[production] |
08:05 |
<Tim> |
killed all image scalers to see if that helps with ms1 load |
[production] |
08:00 |
<Tim> |
killed waiting apache processes |
[production] |
07:35 |
<midom> |
synchronized php-1.5/mc-pmtpa.php |
[production] |
07:24 |
<midom> |
synchronized php-1.5/mc-pmtpa.php 'swapping out srv81' |
[production] |
04:11 |
<Tim> |
fixed /opt/local/bin/zfs-replicate on ms1 to write the snapshot number before starting replication, to avoid permanent error "dataset already exists" after failure |
[production] |
02:16 |
<brion> |
-> https://bugzilla.wikimedia.org/show_bug.cgi?id=19683 |
[production] |
02:12 |
<brion> |
sync-common script doesn't work on nfs-free apaches; language lists etc not being updated. Deployment scripts need to be fixed? |
[production] |
02:03 |
<brion> |
srv159 is absurdly loaded/lagged wtf? |
[production] |
01:58 |
<brion> |
reports of servers with old config, seeing "doesn't exist" for new mhr.wikipedia. checking... |
[production] |
01:16 |
<brion> |
so far so good; CPU graphs on image scalers and ms1 look clean, and I can purge thumbs on commons ok |
[production] |
01:10 |
<brion> |
trying switching image scalers back in for a few, see if they go right back to old pattern or not |
[production] |
01:03 |
<brion> |
load on ms1 has fallen hugely; outgoing network is way up. looks like we're serving out http images fine... of course scaling's dead :P |
[production] |
00:59 |
<brion> |
stopping apache on image scaler boxes, see what that does |
[production] |
00:49 |
<brion> |
attempting to replicate domas's earlier temp success dropping oldest snapshot (last was 4/13): zfs destroy export/upload@weekly-2009-04-20_03:30:00 |
[production] |
00:45 |
<brion> |
restarting nfs server |
[production] |
00:44 |
<brion> |
stopping nfs server, restarting web server |
[production] |
00:40 |
<brion> |
restarting nfs server on ms1 |
[production] |
00:36 |
<brion> |
doesn't seem so far to have changed the NFS access delays on image scalers. |
[production] |
00:31 |
<brion> |
shutting down webserver7 on ms1 |
[production] |
00:23 |
<brion> |
investigating site problem reports. image server stack seems overloaded, so intermittent timeouts on nfs to apaches or http/squid to outside |
[production] |
2009-07-12
§
|
20:30 |
<domas> |
dropped few snapshots on ms1, observed sharp %sys decrease and much better nfs properties immediately |
[production] |
20:05 |
<domas> |
we seem to be hitting issue similar to http://www.opensolaris.org/jive/thread.jspa?messageID=64379 on ms1 |
[production] |
18:55 |
<domas> |
zil_disable=1 on ms1 |
[production] |
18:34 |
<mark> |
Upgraded pybal on lvs3 |
[production] |
18:16 |
<mark> |
Hacked in configurable timeout support for the ProxyFetch monitor of PyBal, set the renderers timeout at 60s |
[production] |
17:58 |
<domas> |
scaler stampedes caused scalers to be depooled by pybal, thus directing stampede to other server in round-robin fashion, all blocking and consuming ms1 SJSWS slots. of course, high I/O load contributed to this. |
[production] |
17:55 |
<domas> |
investigating LVS-based rolling scaler overload issue, Mark and Tim heading the effort now ;-) |
[production] |
17:54 |
<domas> |
bumped up ms1 SJSWS thread count |
[production] |