liu jing created KUDU-3372:
------------------------------
Summary: tserver‘s’ wal directory is too large and still in
increasing
Key: KUDU-3372
URL: https://issues.apache.org/jira/browse/KUDU-3372
Project: Kudu
Issue Type: Bug
Affects Versions: 1.10.0
Reporter: liu jing
We have met a issue that one single tserver's wal directory is too larger
(about larger than 100G), and still in increasing. The such tserver's memory is
also overload, but other tserver is in normal state. So we shut down the
anomaly tserver and delete its wal directory. After rebooting the anomaly
tserver, it recovered.
I have simulated and reproduced the problem. There is the step:
# build a tserver cluster(three nodes).
# set the tserver's hard memory to a low level (about 1GB)
# The client initiates a large number of asynchronous write requests(2 hash
partition, 3 replica, 88 fields)
# Then we will find the tserver that has the leader tablet reproduced the
problem(the directory size can be larger than 4G and the wal segment count can
be larger than 100).
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
