Thanks Mukul!
On 4/12/19 4:11 AM, Mukul Kumar Singh wrote:
Hi Josh,
Following link points to the tweaks done in Ozone for Ratis Server
Please have a look at newRaftPropertiessetRatisLeaderElectionTimeout in
XceiverServerRatis.
https://github.com/apache/hadoop/blob/d31c86892e0ceec5d642f76fc9123fac4fd80db8/hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/transport/server/ratis/XceiverServerRatis.java
setRatisLeaderElectionTimeout sets the leader election timeout in Ozone
in the above file.
Right now, we set it at 1s for leader election timeout.
Thanks,
Mukul
On 4/9/19 6:43 AM, Josh Elser wrote:
Hi,
I'm playing around with Rajeshbabu's VerificationTool for the
LogService and finding that when I introduce more than a naive amount
of load, the daemons are spending most of their time failing to elect
a leader.
Granted, I'm doing testing via Docker that has 8G dedicated to it w/
2G of RAM given to the Java processes, but I feel like I'm not
straying terrible far into craziness (yet).
I'm reading through RaftServerConfigKeys, but I'm not sure which of
these I should be trying to tweak. It's also difficult because there's
so much Ratis logging being thrown to me at INFO, I'm not sure which
is important and which is just garbage -- maybe a follow-up question
in there: what class(es) logging should I care about?
Thanks in advance.
- Josh
