[
https://issues.apache.org/jira/browse/HBASE-11288?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17118793#comment-17118793
]
Duo Zhang commented on HBASE-11288:
-----------------------------------
{quote}
For master local region, the master was light load so we use small heap for
master. And deploy many master on one share machine. There may be many locate
request when many client bootstrap or failover? This is a challange for master.
In our production use case , we meet one case is too many regionserver report
slow down master... Another problem is the availability of master. In the past,
master didn't response user access request so the lower availability is not a
big deal. We can restart master at any time. In a big cluser, the master
bootstrap may take almost 1 minute to load all things from zookeeper or meta
table or receive the report from regionserver. But if master serve the root
table, it may need higher availability.
{quote}
HA master will be good but I think a cache server is enough here. If we want
10k regions per meta, and then 10k metas per root, we can hold 100M regions per
a cluster, which is far beyond our 1M regions per cluster goal. But even for
10k meta regions in root table, typically a row is only several hundreds of
bytes, so for 10k regions, the whole root table is only less then 10MB, which
is not a big deal to pull it every several seconds, and also not a big deal for
the cache server to cache it all in memory.
Of course, 100M regions will make master startup slow as now we need to load
all the rows in meta table when initializing AssignmentManager. We should find
a way to speed it up in the future, but I'd say that this is not a probem for
locating meta region through root table. You can see the patch in HBASE-24389,
the locating meta methods will be available after we finish loading the rows in
root table, before we actually loading the rows in meta table, so it will not
be effect by the slow meta scanning.
> Splittable Meta
> ---------------
>
> Key: HBASE-11288
> URL: https://issues.apache.org/jira/browse/HBASE-11288
> Project: HBase
> Issue Type: Umbrella
> Components: meta
> Reporter: Francis Christopher Liu
> Assignee: Francis Christopher Liu
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
