[
https://issues.apache.org/jira/browse/HDDS-17314?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Siyao Meng updated HDDS-17314:
------------------------------
Description:
{{hdds.heartbeat.recon.interval}} (default 60s) does not control how often a
Datanode heartbeats Recon. {{RunningDatanodeState}} submits a heartbeat to
every endpoint, Recon included, on each SCM heartbeat cycle
({{hdds.heartbeat.interval}}). HDDS-5862 only used the Recon interval as the
timeout for waiting on the Recon heartbeat task, not as its interval, and
HDDS-12639 (a logging change) dropped even that use, apparently by accident. So
Recon has always received heartbeats as often as SCM.
Proposed fix:
* Skip the Recon heartbeat in {{RunningDatanodeState}} until
{{hdds.heartbeat.recon.interval}} has passed since the last one. Version and
register calls to Recon are not delayed.
* Bound the interval: at least {{hdds.heartbeat.interval}}, and at most a third
of {{ozone.scm.stale.node.interval}}, so Recon does not mark Datanodes stale or
dead between heartbeats.
* Set a short Recon interval in the compose clusters that run Recon tests and
in {{MiniOzoneCluster}}, so those tests do not wait longer.
was:
{{hdds.heartbeat.recon.interval}} (default 60s) does not control how often a
Datanode heartbeats Recon. {{RunningDatanodeState}} submits a heartbeat to
every endpoint, Recon included, on each SCM heartbeat cycle
({{hdds.heartbeat.interval}}). The Recon interval is read into {{StateContext}}
(HDDS-5862), but nothing has used it since HDDS-12639 dropped it from the task
wait, so Recon gets heartbeats as often as SCM does.
Proposed fix:
* Skip the Recon heartbeat in {{RunningDatanodeState}} until
{{hdds.heartbeat.recon.interval}} has passed since the last one. Version and
register calls to Recon are not delayed.
* Bound the interval: at least {{hdds.heartbeat.interval}}, and at most a third
of {{ozone.scm.stale.node.interval}}, so Recon does not mark Datanodes stale or
dead between heartbeats.
* Set a short Recon interval in the compose clusters that run Recon tests and
in {{MiniOzoneCluster}}, so those tests do not wait longer.
> Make hdds.heartbeat.recon.interval control the Datanode to Recon heartbeat
> interval
> -----------------------------------------------------------------------------------
>
> Key: HDDS-17314
> URL: https://issues.apache.org/jira/browse/HDDS-17314
> Project: Apache Ozone
> Issue Type: Improvement
> Components: Ozone Datanode, Ozone Recon
> Reporter: Siyao Meng
> Assignee: Siyao Meng
> Priority: Major
> Labels: pull-request-available
>
> {{hdds.heartbeat.recon.interval}} (default 60s) does not control how often a
> Datanode heartbeats Recon. {{RunningDatanodeState}} submits a heartbeat to
> every endpoint, Recon included, on each SCM heartbeat cycle
> ({{hdds.heartbeat.interval}}). HDDS-5862 only used the Recon interval as the
> timeout for waiting on the Recon heartbeat task, not as its interval, and
> HDDS-12639 (a logging change) dropped even that use, apparently by accident.
> So Recon has always received heartbeats as often as SCM.
> Proposed fix:
> * Skip the Recon heartbeat in {{RunningDatanodeState}} until
> {{hdds.heartbeat.recon.interval}} has passed since the last one. Version and
> register calls to Recon are not delayed.
> * Bound the interval: at least {{hdds.heartbeat.interval}}, and at most a
> third of {{ozone.scm.stale.node.interval}}, so Recon does not mark Datanodes
> stale or dead between heartbeats.
> * Set a short Recon interval in the compose clusters that run Recon tests and
> in {{MiniOzoneCluster}}, so those tests do not wait longer.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
