Details
-
Bug
-
Status: Resolved
-
Blocker
-
Resolution: Fixed
-
1.3.0
Description
This change seems to have caused a bad file handle leak on the datanodes:
Reverting it, causes the problem to go away. You can reproduce the problem like so:
cd hadoop-ozone/dist/target/ozone-1.3.0-SNAPSHOT/compose/ozone
docker-compose up --no-recreate --scale datanode=3 -d
Then watch the file handle count on a datanode:
docker exec -it ozone_datanode_1 bash
sudo bash
yum install -y lsof
while : ; do echo -n "$(date) open files: "; lsof -u hadoop | wc -l; sleep 3s; done
Meanwhile run freon from a different node:
docker exec -it ozone_om_1 bash
ozone freon ockg --volume=btest --bucket=freon --threads=24 --replication-type=RATIS --factor=THREE --size=1024 --number-of-tests=2000000 --prefix=small-file
and the datanode will look like this within a few minutes:
Thu Jul 21 22:42:45 UTC 2022 open files: 954
Thu Jul 21 22:42:48 UTC 2022 open files: 2184
Thu Jul 21 22:42:51 UTC 2022 open files: 3843
Thu Jul 21 22:42:54 UTC 2022 open files: 5838
Thu Jul 21 22:42:58 UTC 2022 open files: 8240
Thu Jul 21 22:43:01 UTC 2022 open files: 11827
Thu Jul 21 22:43:04 UTC 2022 open files: 16641
Thu Jul 21 22:43:08 UTC 2022 open files: 21931
Thu Jul 21 22:43:11 UTC 2022 open files: 27470
Thu Jul 21 22:43:15 UTC 2022 open files: 33239
Thu Jul 21 22:43:19 UTC 2022 open files: 39027
Thu Jul 21 22:43:23 UTC 2022 open files: 45251
Thu Jul 21 22:43:27 UTC 2022 open files: 51667
Thu Jul 21 22:43:31 UTC 2022 open files: 57262
Attachments
Issue Links
- is caused by
-
HDDS-6794 EC: Analyze and add putBlock even on non writing node in the case of partial single stripe.
- Resolved
- links to