You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues-all@impala.apache.org by "Alexey Serbin (Jira)" <ji...@apache.org> on 2022/02/25 19:51:00 UTC
[jira] [Commented] (IMPALA-11154) Idle Kudu daemons consume too much CPU
[ https://issues.apache.org/jira/browse/IMPALA-11154?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17498286#comment-17498286 ]
Alexey Serbin commented on IMPALA-11154:
----------------------------------------
I guess the CPU usage is due to the Raft heartbeating between tablet replicas. In Kudu, idle tablet replicas send and process Raft heartbeats.
These JIRA items might be useful to get more context:
* https://issues.apache.org/jira/browse/KUDU-3134
* https://issues.apache.org/jira/browse/KUDU-1973
> Idle Kudu daemons consume too much CPU
> --------------------------------------
>
> Key: IMPALA-11154
> URL: https://issues.apache.org/jira/browse/IMPALA-11154
> Project: IMPALA
> Issue Type: Bug
> Components: Infrastructure
> Reporter: Zoltán Borók-Nagy
> Assignee: Zoltán Borók-Nagy
> Priority: Major
> Attachments: perf_record_1.svg, perf_record_kudu.svg
>
>
> I noticed that if I issue {{top}} in a machine that runs the minicluster, Kudu is always amongst the highest CPU consumers:
> {noformat}
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 29730 borokna+ 20 0 1438712 101340 29256 S 27.8 0.3 17:43.76 kudu-tserver
> 29682 borokna+ 20 0 1435780 100764 29360 S 16.7 0.3 16:24.49 kudu-tserver
> 29700 borokna+ 20 0 1442964 104636 29452 S 16.7 0.3 18:18.77 kudu-tserver
> {noformat}
> The above is for a Kudu cluster that virtually doesn't do any work.
> The flagfile looks like the following:
> {noformat}
> -rpc_bind_addresses=localhost:31202
> -webserver_port=31302
> -fs_wal_dir=/home/boroknagyz/Impala/testdata/cluster/cdh7/node-1/var/lib/kudu/ts/wal
> -fs_data_dirs=/home/boroknagyz/Impala/testdata/cluster/cdh7/node-1/var/lib/kudu/ts/data
> -log_dir=/home/boroknagyz/Impala/testdata/cluster/cdh7/node-1/var/log/kudu/ts
> # The flags below require unsafe flags to be unlocked.
> -unlock_unsafe_flags
> # fsync is disabled for additional speed. Sometimes operations are slow on EC2/GCE test
> # machines. Some data loss could occur if the system crashes before the OS has a chance
> # to flush data to disk but that is acceptable for development purposes.
> -never_fsync
> # There is no need to require NTP-synchronized clock for tests where all the
> # participating Kudu masters and tablet servers are run at the same node using the same
> # local wallclock.
> -time_source=system_unsync
> # Enable Kudu transaction.
> -enable_txn_system_client_init
> -unlock_experimental_flags
> {noformat}
> Kudu version info is
> {noformat}
> kudu 67ba3cae45
> revision 67ba3cae45143d083bdac119f45278ce0d2bfefb-dirty
> build type RELEASE
> built by None at 11 Jan 2022 01:08:09 UTC on 4cf70af93dc4
> {noformat}
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org