[
https://issues.apache.org/jira/browse/MESOS-2110?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14605408#comment-14605408
]
Adam B commented on MESOS-2110:
-------------------------------
Committed the feature. There's still a pending review to update the tests:
https://reviews.apache.org/r/35958/
commit 6cd28dd9154a1a28ff0756078cb776c221d025d1
Author: Adam B <[email protected]>
Date: Mon Jun 29 02:59:59 2015 -0700
Added Configurable Slave Ping Timeouts
Added new --slave_ping_timeout and --max_slave_ping_timeouts flags
to mesos-master to supplement the DEFAULT_SLAVE_PING_TIMEOUT (15secs)
and DEFAULT_MAX_SLAVE_PING_TIMEOUTS (5).
These can be extended if slaves are expected/allowed to be down for
longer than a minute or two.
Slave will receive master's ping timeout in SlaveRe[re]gisteredMessage.
Beware that this affects recovery from network timeouts as well as
actual slave node/process failover.
Also fixed the log message in recoveredSlavesTimeout() to correctly
reference flags.slave_reregister_timeout instead of the unrelated
ping timeouts.
Review: https://reviews.apache.org/r/29507
> Configurable Ping Timeouts
> --------------------------
>
> Key: MESOS-2110
> URL: https://issues.apache.org/jira/browse/MESOS-2110
> Project: Mesos
> Issue Type: Improvement
> Components: master, slave
> Reporter: Adam B
> Assignee: Adam B
> Labels: master, mesosphere, network, slave, timeout
>
> After a series of ping-failures, the master considers the slave lost and
> calls shutdownSlave, requiring such a slave that reconnects to kill its tasks
> and re-register as a new slaveId. On the other side, after a similar timeout,
> the slave will consider the master lost and try to detect a new master. These
> timeouts are currently hardcoded constants (5 * 15s), which may not be
> well-suited for all scenarios.
> - Some clusters may tolerate a longer slave process restart period, and
> wouldn't want tasks to be killed upon reconnect.
> - Some clusters may have higher-latency networks (e.g. cross-datacenter, or
> for volunteer computing efforts), and would like to tolerate longer periods
> without communication.
> We should provide flags/mechanisms on the master to control its tolerance for
> non-communicative slaves, and (less importantly?) on the slave to tolerate
> missing masters.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
