uber / cherami-client-go / 194 / 1
81%
master: 83%

LAST BUILD BRANCH: v2.6.0
DEFAULT BRANCH: master
Ran
Files 5
Run time 0s
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 80.903%. First build
194.1

push

travis-ci

web-flow 
Unblock read acks pump from inputhost (#17)

* Unblock read acks pump from inputhost

If the server is draining gracefully, then the server will send a
DRAIN command. The DRAIN command used to simply stop the write pump and
wait for the server to close the stream.

Even though server successfully will close the stream after finishing
the DRAIN, the readAcks pump will never see the EOF because we don't do
a stream.Read() unless we have some messages inflight.

This patch tries to solve that by waiting for a default of a minute and
explicitly closing the connection. In addition this patch also adds some
additional logs and metrics to make sure we can track retries and
failures on publish.

1. We need to make sure if we are already closing/draining we shouldn't
close the connection. This is critical because during reconfig we
could decide to close the connection but in the meanwhile we could
have received a drain command and started draining. In that case we
should just allow the drain process to take care of the rest.
2. Instead of waiting for the entire timeout period, we can ideally just
check for the number of responses we received and bail out immediately.

Fix connection close to make sure we always wait for sometime to give
a chance to drain.

We can avoid using the atomic variables for closed and we can just rely
on the shutting down channel to make sure we close is idempotent.

* Make sure we wait for the final drain timeout

If the checkDrainTimer fires, we exit immediately even if the
drain has not finished yet. But we should wait for the bigger
timeout as well.

233 of 288 relevant lines covered (80.9%)

736.15 hits per line

Source Files on job 194.1