hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ray Chiang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HADOOP-12664) UGI auto-renewer does not verify kinit availability during initialization
Date Mon, 09 Jan 2017 19:43:58 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-12664?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15812650#comment-15812650

Ray Chiang commented on HADOOP-12664:

I'm pretty agnostic with respect to whatever solution we end up choosing.  At the very least,
I see the following errors:

1) kinit not found (not installed, configuration problem)
2) kinit temporarily not accessible (e.g. flaky filesystem)
3) kinit fails to renew intermittently or has slow response/timeouts (e.g. network)

Currently, there are two possible checks:

A) Do the kinit path search
B) Simply check kinit exit code 127 for "command not found"

G) Do check outside the kinit retry thread
H) Do check within the kinit retry thread

Given potentially 1) and 2) getting conflated, if we choose option B) and H),
we might want to set some kind of consecutive "command not found" retry error
threshold and throwing an exception if we exit based on that situation.


> UGI auto-renewer does not verify kinit availability during initialization
> -------------------------------------------------------------------------
>                 Key: HADOOP-12664
>                 URL: https://issues.apache.org/jira/browse/HADOOP-12664
>             Project: Hadoop Common
>          Issue Type: Improvement
>            Reporter: Arpit Agarwal
>            Assignee: Ray Chiang
>            Priority: Minor
>              Labels: supportability
>         Attachments: HADOOP-12664.001.patch, HADOOP-12664.002.patch, HADOOP-12664.003.patch
> UGI auto-renewer does not verify that {{hadoop.kerberos.kinit.command}} is in the path
during initialization. If not available, the auto-renewal thread will hit an error during
TGT renewal. We recently saw a case where it manifests as transient errors during client program
execution which can be hard to track down without UGI logging.
> It seems like {{kinit}} availability should be verified during initialization to make
the behavior more predictable.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org

View raw message