Ottomata has uploaded a new change for review.

  https://gerrit.wikimedia.org/r/163669

Change subject: Reset queue_buffering_max_ms to default 1000 ms for varnishkafka
......................................................................

Reset queue_buffering_max_ms to default 1000 ms for varnishkafka

We increased this in attempt to avoid produce errors during
kafka <-> zookeeper timeouts.  However, this just caused
the number of buffered messages in varnishkafka to increase,
but did not actually help wtih produce errors.  As far as I can
tell currently, varnishkafka and librdkafka are not responsible
for lost messages during zookeeper timeouts.

http://cl.ly/image/3N3i2L04161r shows the msgq_cnt buffer
increase on esams bits since I started tweaking this setting.

You can read more about the bug we are trying to figure out here:
https://bugzilla.wikimedia.org/show_bug.cgi?id=69667

Change-Id: If6a9d5e58c2917f2e37666ff6f70c44f2526ec15
---
M manifests/role/cache.pp
1 file changed, 1 insertion(+), 6 deletions(-)


  git pull ssh://gerrit.wikimedia.org:29418/operations/puppet 
refs/changes/69/163669/1

diff --git a/manifests/role/cache.pp b/manifests/role/cache.pp
index 526b16a..b02a0fb 100644
--- a/manifests/role/cache.pp
+++ b/manifests/role/cache.pp
@@ -521,12 +521,7 @@
                 # At ~6000 msgs per second, 500000 messages is over 1 minute
                 # of buffering, which should be more than enough.
                 queue_buffering_max_messages => 500000,
-                # Broker <-> Zookeeper timeout is currently set at
-                # 16 seconds, and it can take a few seconds for leadership
-                # change metadata to propagate to down to varnishkafka.
-                # Setting this to 30 seconds.
-                queue_buffering_max_ms       => 30000,
-                # bits varnishes do about 6000 reqs / sec each.
+                # bits varnishes can do about 6000 reqs / sec each.
                 # We want to send batches at least once a second.
                 batch_num_messages           => 6000,
                 # large timeout to account for potential cross DC latencies

-- 
To view, visit https://gerrit.wikimedia.org/r/163669
To unsubscribe, visit https://gerrit.wikimedia.org/r/settings

Gerrit-MessageType: newchange
Gerrit-Change-Id: If6a9d5e58c2917f2e37666ff6f70c44f2526ec15
Gerrit-PatchSet: 1
Gerrit-Project: operations/puppet
Gerrit-Branch: production
Gerrit-Owner: Ottomata <o...@wikimedia.org>

_______________________________________________
MediaWiki-commits mailing list
MediaWiki-commits@lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/mediawiki-commits

Reply via email to