Shenandoah SIGSEGV crashes on jdk8 build

[Morten Vejen Nielsen]

Hi,

I'm not sure if this is the right place to post this, if it isn't I
apologize in advance.


We are currently using Shenandoah with Cassandra 3.11 for multi-tenant
platform. This is run with the jdk-8 version of the redhat jdk that
includes Shenandoah. We are fully aware that this in itself is an
experimental setup but we found this to be the best solution to mitigate
issues with high gc pauses and request timeouts in cassandra.

So far we have had really good experience with this, but we have
unfortunately seen a few crashes now where the JVM crashes with a SIGSEGV
(0xb) during a high pressure situation.

I have attached a hs_err_pid file with the detailed error, unfortunately I
don't have a core dump at the moment but we are currently trying to
reproduce this error consistently to attain this.

At the time it crashed (10:30) we noticed high CPU activity on the system
and high GC activity (have attached gc log file). I noticed this logging in
gc log, which to me seems to indicate that our gc cant keep up with
application allocation rate, so we are currently looking at tuning this.

2021-01-20T10:50:28.918+0000: 4063718.351: Total time for which application
threads were stopped: 0.0007702 seconds, Stopping threads took: 0.0000455
seconds
Trigger: Average GC time (1201.36 ms) is above the time for allocation rate
(739 MB/s) to deplete free headroom (874M)

But it seems to me that it shouldn't crash with a SIGSEGV because of this.

So I wanted to post this message in case to ask if this is a known issue or
if this is a bug?

We are using this jdk 8 build:
build
1.8.0-builds.shipilev.net-openjdk-shenandoah-jdk8-b766-20201130-testing-b00

And otherwise we are running with default gc settings except for having set:
-XX:ParallelGCThreads=16
-XX:ConcGCThreads=5

Best regards
Morten