uber / cherami-server / 1178
67%

DEFAULT BRANCH: master
Ran
Jobs 1
Files 128
Run time 7s
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
1178

cron

travis-ci

web-flow 
Graceful drain on inputhost along with client drain (#112)

* Add graceful drain of extents on inputhost

This patch simply implements the graceful drain on inputhost.
Whenever an inputhost gets a graceful drain request, it
does the following:
(1) Marks the extent as draining and stops the writepump and waits
    for the write pump to drain (for a timeout period)
(2) On the read pump if we have drained all the messages, then we
    can safely close the extent and go away.

A couple of caveats:
1. Note: close is idempotent and so we should be able to force close
   an extent anytime during the drain process.
2. We just need the RLock() during read pump aggregation and since it
   is not contended the profiling of inputhost didn't reveal it to be
   at top.

* Add support for draining client connections
  * To make sure we do this correctly, we maintain a "WritableExtents"
  metric which is slightly different from the "OpenExtents" because
  a drained extent is still open but is read only.
  * When the number of WritableExtents goes to zero, we should send a
  DRAIN command on all the connections.
  * The connections will simply wait for a timeout period for the drain to
  finish. If things finish successfully, we will unload the pathCache
  and close all connections automatically. If not, the timeout period will
  hit and we will drop the connection object.
  * Add an integration test to test this flow.

* Make drain extent to return synchronously

* Update to pull recent client side changes

19895 of 28802 relevant lines covered (69.08%)

2092.46 hits per line

Jobs
ID Job ID Ran Files Coverage
1 1178.1 128
69.08
 Travis Job 1178.1
Source Files on build 1178
Coverage ∆ File Lines Relevant Covered Missed Hits/Line