[
https://issues.apache.org/jira/browse/HBASE-13260?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14506155#comment-14506155
]
Enis Soztutar commented on HBASE-13260:
---------------------------------------
bq. try to pass 50 as the WAL threads number (there is just 1 thread in the wal)
Let me try that.
bq. but still, I don't think the focus should be comparison on write perf
write perf comes into picture in case we will do assignments using procedures.
In case of 100K/1M region clusters, we do not want to be bottlenecked by the
proc store.
bq. can we throw away the logs when we remove everything from the memstore?
I am not sure I follow. Rolling the WAL is a very heavyweight operation. We
should not roll the WAL everytime there is no more procedure references. We
already have the region mechanics to optimize the WAL rolls vs flushes. We also
have the periodic flushes to help with recovery (although the patch does not
contain periodic flusher yet).
bq. can we avoid the delete markers and similar.
I think this would be a bigger change. We want if memstore contains it remove,
otherwise put a tombstone to optimize empty flushes (for this logic).
Regardless of this, do you think flushes and compactions will be a perf
problem? As I comment above, every time a flush happens, we will get rid of the
WALs using the regular seqId tracking. Better yet, on graceful shutdown, we
will not do any replay, since everything is flushed and WAL files already
archived.
One of the goals for this patch is not to just replace proc store, but have
this as a general service for other use cases (storing cluster membership, RS
groups, etc). The motivation is to also not have yet another WAL format. In
case we do not want the rest of HRegion, I think we should at least use FSHLog
as the proc wal store.
> Bootstrap Tables for fun and profit
> ------------------------------------
>
> Key: HBASE-13260
> URL: https://issues.apache.org/jira/browse/HBASE-13260
> Project: HBase
> Issue Type: Bug
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Fix For: 2.0.0, 1.1.0
>
> Attachments: hbase-13260_bench.patch, hbase-13260_prototype.patch
>
>
> Over at the ProcV2 discussions(HBASE-12439) and elsewhere I was mentioning an
> idea where we may want to use regular old regions to store/persist some data
> needed for HBase master to operate.
> We regularly use system tables for storing system data. acl, meta, namespace,
> quota are some examples. We also store the table state in meta now. Some data
> is persisted in zk only (replication peers and replication state, etc). We
> are moving away from zk as a permanent storage. As any self-respecting
> database does, we should store almost all of our data in HBase itself.
> However, we have an "availability" dependency between different kinds of
> data. For example all system tables need meta to be assigned first. All
> master operations need ns table to be assigned, etc.
> For at least two types of data, (1) procedure v2 states, (2) RS groups in
> HBASE-6721 we cannot depend on meta being assigned since "assignment" itself
> will depend on accessing this data. The solution in (1) is to implement a
> custom WAL format, and custom recover lease and WAL recovery. The solution in
> (2) is to have the table to store this data, but also cache it in zk for
> bootrapping initial assignments.
> For solving both of the above (and possible future use cases if any), I
> propose we add a "boostrap table" concept, which is:
> - A set of predefined tables hosted in a separate dir in HDFS.
> - A table is only 1 region, not splittable
> - Not assigned through regular assignment
> - Hosted only on 1 server (typically master)
> - Has a dedicated WAL.
> - A service does WAL recovery + fencing for these tables.
> This has the benefit of using a region to keep the data, but frees us to
> re-implement caching and we can use the same WAL / Memstore / Recovery
> mechanisms that are battle-tested.
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
