accumulo-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Josh Elser (JIRA)" <>
Subject [jira] [Commented] (ACCUMULO-4458) lock contention around configuration settings impacts tablet server performance
Date Thu, 15 Sep 2016 15:11:21 GMT


Josh Elser commented on ACCUMULO-4458:

I think this is OK as-is [~busbey], but I'm still left wondering:

bq. It seems like it would be more simple if we just extracted all Accumulo properties (instead
of this reduced list). Is there a big impact on memory for this?

I just can't get over a worry that the HOT_PATH_PROPERTIES will rot and your optimization
will be undone (or less efficient). We don't have _that_ many properties, I think it might
be better long-term to just {{put(prop.getKey(), null)}} for all of the Property's that are
not explicitly defined.

bq. Forgot to mention above that the set method (that we effectively ignore for any configs
that have been set in the xml configs) is expressly labeled as something that's only for test,
which is why I think it's fine to ignore it at runtime.
bq. Any calls to set won't be reflected in staticConfigs, because they're read-only after
instantiation of SiteConfiguration so that concurrent access can be safe.

Agreed. Makes me think that we could do better at an API which allows injection of properties
in some TestConfiguration (instead of trying to shim things in via {{set()}}). Maybe my (immediately)
above worry could be alleviated by better test-API and then simplification of things here?
(e.g. follow-on)

> lock contention around configuration settings impacts tablet server performance
> -------------------------------------------------------------------------------
>                 Key: ACCUMULO-4458
>                 URL:
>             Project: Accumulo
>          Issue Type: Bug
>          Components: tserver
>    Affects Versions: 1.7.1, 1.8.0
>            Reporter: Sean Busbey
>            Assignee: Sean Busbey
>            Priority: Critical
>             Fix For: 1.7.3, 1.8.1, 2.0.0
>         Attachments: ACCUMULO-4458-1.7.v1.patch
> While investigating a pretty severe performance regression comparing YCSB against 1.6
and 1.7, I found a fair bit of lock contention around getting configuration values. This was
improved by ACCUMULO-4388, but various threads eventually all started contending on the configuration
values expected from the site xml files.

This message was sent by Atlassian JIRA

View raw message