clemlabprojects / ambari

Fork of Apache Ambari maintained by Clemlab Company
https://www.clemlab.com
Apache License 2.0
44 stars 19 forks source link

odp 1.2.4.0 / Oracle linux 9 / Python 3 `curl: option --data-binary: out of memory` #105

Open zorel opened 1 day ago

zorel commented 1 day ago

odp 1.2.4.0 / Oracle linux 9 / Python 3

https://github.com/curl/curl/issues/1385 Hit curl: option --data-binary: out of memory when trying to start Yarn ResourceManager on this command line:

Execution of \'curl -sS -L -w \'%{http_code}\' -X PUT --data-binary @/var/lib/ambari-agent/tmp/yarn-ats/1.2.4.0-77/hbase.tar.gz -H \'Content-Type: application/octet-stream\' \'http://control-001:50070/webhdfs/v1/odp/apps/1.2.4.0-77/hbase/hbase.tar.gz?op=CREATE&overwrite=True&permission=444&user.name=hdfs\' 1>/tmp/tmp0shcomog 2>/tmp/tmpu3893ebt\' returned 2

Running manually with -T worked:

curl -sS -L -w '%{http_code}' -X PUT -T /var/lib/ambari-agent/tmp/yarn-ats/1.2.4.0-77/hbase.tar.gz -H 'Content-Type: application/octet-stream' 'http://control-001:50070/webhdfs/v1/odp/apps/1.2.4.0-77/hbase/hbase.tar.gz?op=CREATE&overwrite=True&permission=444&user.name=hdfs'
lucasbak commented 1 day ago

@zorel

How much of memory do you have on your node (from where the curl is done ?) and also how much memory heap did you configured for your namenode ?

Regards,

zorel commented 1 day ago

Enough on the node, but indeed the NN is configured with 1GB only (which is the default at installation time). But interestingly, the message comes from curl, not the NN process.

zorel commented 1 day ago

/var/lib/ambari-agent/tmp/yarn-ats/1.2.4.0-77/hbase.tar.gz is 1.1GB

lucasbak commented 1 day ago

It would be interesting to try again with higher memory on the NN like 2 or 4Go

Regards

zorel commented 20 hours ago

Indeed it works with higher heaps. 2GB works.

lucasbak commented 20 hours ago

Hi @zorel,

Thank you ;) It may be linked to JDK 8 version (compared to old JDK 6/7 when the service advisor has been written) https://github.com/clemlabprojects/ambari/blob/c4f200b7994ff724e445cabaff5eaaa5367537b0/ambari-server/src/main/resources/stacks/ODP/1.0/services/HDFS/service_advisor.py#L199 is where the 1024 is set, we can update it to 2048

Regards

zorel commented 17 hours ago

It does not work. I made the test on ambari-server node, and instead of a error cause the hbase.tar.gz does not exists on this node, curl does not complains and upload an emtpy file. heap size does not solve the issue.