[
https://issues.apache.org/jira/browse/ACCUMULO-3022?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14077744#comment-14077744
]
Eric Newton commented on ACCUMULO-3022:
---------------------------------------
There's a limit on the number of zookeeper watches (see ACCUMULO-2757), and
watches grow with the size of the cluster * the number tables * the number of
configuration items per table.
When the number of watchers set by clients grows to 100M or so on today's
hardware (10's of Gs of ram), we should begin to worry. I would argue that on
large clusters (>1000), that limits the number of tables to ~500-1000, which
seems reasonable.
The size of a table name is not a big problem... so long as it stays under a
megabyte or so. Sanity checking is useful to prevent user errors from tanking
a whole zookeeper quorum.
> Limit size of data objects in ZooKeeper
> ---------------------------------------
>
> Key: ACCUMULO-3022
> URL: https://issues.apache.org/jira/browse/ACCUMULO-3022
> Project: Accumulo
> Issue Type: Improvement
> Affects Versions: 1.4.0
> Reporter: Keith Turner
> Priority: Minor
> Fix For: 1.7.0
>
>
> This issue is inspired by ACCUMULO-3021. The original issue was concerning
> table names, but I've expanded the issue to include other data objects in
> ZooKeeper (see comments below), especially in light of the fact that there's
> already an issue specifically addressing giant table names (ACCUMULO-2366).
--
This message was sent by Atlassian JIRA
(v6.2#6252)
