[
https://issues.apache.org/jira/browse/HIVE-18097?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-18097:
------------------------------------
Description:
Final design TBD.
We may add an option to move triggers: "on full kill/ignore/delay kill".
The former will kill the query if the dest is full.
The ignore one would not perform the move... The thinking is that admin would
add another trigger to kill queries after some time (e.g. - move after 10s,
kill after 30s), and thus give slow/large queries a grace period in case dest
is full so they cannot gracefully release the capacity.
It's possible to add a third option - "delay kill", where we would ignore the
move, but kill the query if someone requests capacity and the pool is full.
Note that none of these requires oversubscription which is a pain to track.
was:
Final design TBD.
We may add an option to the trigger: "on full kill/ignore/delay kill".
The former will kill the query if the dest is full.
The ignore one would not perform the move... The thinking is that admin would
add another trigger to kill queries after some time (e.g. - move after 10s,
kill after 30s), and thus give slow/large queries a grace period in case dest
is full so they cannot gracefully release the capacity.
It's possible to add a third option - "delay kill", where we would ignore the
move, but kill the query if someone requests capacity and the pool is full.
Note that none of these requires oversubscription which is a pain to track.
> query move phase 2 - handling the destination pool being full
> -------------------------------------------------------------
>
> Key: HIVE-18097
> URL: https://issues.apache.org/jira/browse/HIVE-18097
> Project: Hive
> Issue Type: Sub-task
> Reporter: Sergey Shelukhin
>
> Final design TBD.
> We may add an option to move triggers: "on full kill/ignore/delay kill".
> The former will kill the query if the dest is full.
> The ignore one would not perform the move... The thinking is that admin would
> add another trigger to kill queries after some time (e.g. - move after 10s,
> kill after 30s), and thus give slow/large queries a grace period in case dest
> is full so they cannot gracefully release the capacity.
> It's possible to add a third option - "delay kill", where we would ignore the
> move, but kill the query if someone requests capacity and the pool is full.
> Note that none of these requires oversubscription which is a pain to track.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
