In some of my testing on HDFS 2.6.0 and agitation, I believe leases
weren't being recovered until the datanode was restarted (which
agitation killed). Every time I jstack'ed the master, I saw it hung on
the HDFS recoverLease() method. I saw nothing that suggested a bug in
Accumulo (a nod towards Christopher's assessment).
Eric Newton wrote:
Can you quantify "Some lease recoveries taking a long time in HDFS."?
I think Keith and I are ok with the release. We're looking into our
(shared) testing environment to figure out why Accumulo and HDFS are not
getting along during recoveries.
-Eric
On Fri, May 15, 2015 at 3:51 PM, Josh Elser<josh.el...@gmail.com> wrote:
72hr 3node CI verify w/ agitation just succeeded.
org.apache.accumulo.test.continuous.ContinuousVerify$Counts
REFERENCED=19706560041
UNREFERENCED=29844289
Repeatedly ran into issues with HoldTime from HDFS (wrote a loop to just
restart the ingester). Some lease recoveries taking a long time in HDFS.
Nothing super impactful.
I believe that wraps up all required testing.
Josh Elser wrote:
24hr CI just verified
org.apache.accumulo.test.continuous.ContinuousVerify$Counts
REFERENCED=5754999773
UNREFERENCED=6000227
Josh Elser wrote:
24hr RW on 3 nodes w/ agitation just finished positively.
Josh Elser wrote:
Testing: All unit and integration tests are passing. Completed 3-day CI
w/o agitation or verification. Completed 24hr RandomWalk w/o agitation
on 3 nodes. 95% through 24hr CI and RW w/ agitation.
