[
https://issues.apache.org/jira/browse/HBASE-14468?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15084059#comment-15084059
]
Lars Hofhansl commented on HBASE-14468:
---------------------------------------
That's different. I see that as I flush at the time of flushing we do
compaction checks and remove the old files.
But if I write a bunch in 30s (I flush manually), and then stop all activity
the files are not purged.
With longer TTLs of a few days or weeks that may significant.
I tested this is the shell with:
{code}
put 'x', 'r1', 'y:1', 1; flush 'x'
{code}
And just changing 'r1' to 'r2', 'r3', etc, each time. That way I created 6
store files in 30s, and they never get collected, even though their TTL clearly
expires. (compaction threshold is still 3). With the default collector they do
get collected after a minute or so.
Is there something else besides this patch I need that is 1.2+ but not in 0.98?
> Compaction improvements: FIFO compaction policy
> -----------------------------------------------
>
> Key: HBASE-14468
> URL: https://issues.apache.org/jira/browse/HBASE-14468
> Project: HBase
> Issue Type: Improvement
> Components: Compaction, Performance
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
> Fix For: 2.0.0, 1.2.0, 1.3.0
>
> Attachments: 14468-0.98-v2.txt, 14468-0.98.txt, HBASE-14468-v1.patch,
> HBASE-14468-v10.patch, HBASE-14468-v2.patch, HBASE-14468-v3.patch,
> HBASE-14468-v4.patch, HBASE-14468-v5.patch, HBASE-14468-v6.patch,
> HBASE-14468-v7.patch, HBASE-14468-v8.patch, HBASE-14468-v9.patch,
> HBASE-14468.add.patch
>
>
> h2. FIFO Compaction
> h3. Introduction
> FIFO compaction policy selects only files which have all cells expired. The
> column family MUST have non-default TTL.
> Essentially, FIFO compactor does only one job: collects expired store files.
> These are some applications which could benefit the most:
> # Use it for very high volume raw data which has low TTL and which is the
> source of another data (after additional processing). Example: Raw
> time-series vs. time-based rollup aggregates and compacted time-series. We
> collect raw time-series and store them into CF with FIFO compaction policy,
> periodically we run task which creates rollup aggregates and compacts
> time-series, the original raw data can be discarded after that.
> # Use it for data which can be kept entirely in a a block cache (RAM/SSD).
> Say we have local SSD (1TB) which we can use as a block cache. No need for
> compaction of a raw data at all.
> Because we do not do any real compaction, we do not use CPU and IO (disk and
> network), we do not evict hot data from a block cache. The result: improved
> throughput and latency both write and read.
> See: https://github.com/facebook/rocksdb/wiki/FIFO-compaction-style
> h3. To enable FIFO compaction policy
> For table:
> {code}
> HTableDescriptor desc = new HTableDescriptor(tableName);
>
> desc.setConfiguration(DefaultStoreEngine.DEFAULT_COMPACTION_POLICY_CLASS_KEY,
> FIFOCompactionPolicy.class.getName());
> {code}
> For CF:
> {code}
> HColumnDescriptor desc = new HColumnDescriptor(family);
>
> desc.setConfiguration(DefaultStoreEngine.DEFAULT_COMPACTION_POLICY_CLASS_KEY,
> FIFOCompactionPolicy.class.getName());
> {code}
> Although region splitting is supported, for optimal performance it should be
> disabled, either by setting explicitly DisabledRegionSplitPolicy or by
> setting ConstantSizeRegionSplitPolicy and very large max region size. You
> will have to increase to a very large number store's blocking file number :
> *hbase.hstore.blockingStoreFiles* as well.
>
> h3. Limitations
> Do not use FIFO compaction if :
> * Table/CF has MIN_VERSION > 0
> * Table/CF has TTL = FOREVER (HColumnDescriptor.DEFAULT_TTL)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
