[
https://issues.apache.org/jira/browse/DRILL-4207?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Abhishek Girish updated DRILL-4207:
-----------------------------------
Description:
On a on-demand test environment, which brings up a Hadoop cluster with Drill &
Hive, I observed that queries submitted to Drill appear to hang (but eventually
do complete - after several minutes). This occurs frequently and can be
reproduced.
The web UI shows them to be in a PENDING state and additional queries cannot be
submitted at that time (they too go into PENDING state). These queries also
take a very long time to cancel. Basically the whole Drill setup is in a
unusable state.
Logs indicate that the query takes prolonged time in registering with the Hive
metastore. It could be either the hive metastore hasn't come up completely or
because Drill is having trouble communicating with Hive server. Log snippet
attached.
Disabling Hive helps avoid this issue.
Since the queries being run are not related to Hive and do not need to use the
Hive plugin, I'm not quite sure why Drill would need to connect to Hive in the
first place. And if Drill does requires every query to register with storage
plugins such as Hive, then there should possibly be a timeout if the query is
unrelated to the plugin taking time to load/register.
was:
On a on-demand test environment, which brings up a Hadoop cluster with Drill, I
observed that queries submitted to Drill appear to hang (but eventually do
complete - after several minutes). This occurs frequently and can be reproduced.
The web UI shows them to be in a PENDING state and additional queries cannot be
submitted at that time (they too go into PENDING state). These queries also
take a very long time to cancel. Basically the whole Drill setup is in a
unusable state.
Logs indicate that the query takes prolonged time in registering with the Hive
metastore. It could be either the hive metastore hasn't come up completely or
because Drill is having trouble communicating with Hive server. Log snippet
attached.
Disabling Hive helps avoid this issue.
Since the queries being run are not related to Hive and do not need to use the
Hive plugin, I'm not quite sure why Drill would need to connect to Hive in the
first place. And if Drill does requires every query to register with storage
plugins such as Hive, then there should possibly be a timeout if the query is
unrelated to the plugin taking time to load/register.
> Queries appear to hang in scenarios when hive plugin is enabled
> ---------------------------------------------------------------
>
> Key: DRILL-4207
> URL: https://issues.apache.org/jira/browse/DRILL-4207
> Project: Apache Drill
> Issue Type: Bug
> Components: Query Planning & Optimization, Storage - Hive
> Affects Versions: 1.4.0, 1.5.0
> Reporter: Abhishek Girish
> Priority: Critical
> Attachments: drillbit.log.txt
>
>
> On a on-demand test environment, which brings up a Hadoop cluster with Drill
> & Hive, I observed that queries submitted to Drill appear to hang (but
> eventually do complete - after several minutes). This occurs frequently and
> can be reproduced.
> The web UI shows them to be in a PENDING state and additional queries cannot
> be submitted at that time (they too go into PENDING state). These queries
> also take a very long time to cancel. Basically the whole Drill setup is in a
> unusable state.
> Logs indicate that the query takes prolonged time in registering with the
> Hive metastore. It could be either the hive metastore hasn't come up
> completely or because Drill is having trouble communicating with Hive
> server. Log snippet attached.
> Disabling Hive helps avoid this issue.
> Since the queries being run are not related to Hive and do not need to use
> the Hive plugin, I'm not quite sure why Drill would need to connect to Hive
> in the first place. And if Drill does requires every query to register with
> storage plugins such as Hive, then there should possibly be a timeout if the
> query is unrelated to the plugin taking time to load/register.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
