Apache
/
hbase
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

HBASE-20390: IMC Default Parameters for 2.0.0

This commit is contained in:
eshcar 2018-05-24 18:10:10 +03:00
parent bfab1e2f92
commit 079e08d7c5
4 changed files with 13 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/CompactingMemStore.java
View File

@ -43,12 +43,12 @@ import org.apache.hadoop.hbase.wal.WAL;
/**
* A memstore implementation which supports in-memory compaction.
* A compaction pipeline is added between the active set and the snapshot data structures;
* it consists of a list of kv-sets that are subject to compaction.
* Like the snapshot, all pipeline components are read-only; updates only affect the active set.
* it consists of a list of segments that are subject to compaction.
* Like the snapshot, all pipeline segments are read-only; updates only affect the active set.
* To ensure this property we take advantage of the existing blocking mechanism -- the active set
* is pushed to the pipeline while holding the region's updatesLock in exclusive mode.
* Periodically, a compaction is applied in the background to all pipeline components resulting
* in a single read-only component. The ``old'' components are discarded when no scanner is reading
* Periodically, a compaction is applied in the background to all pipeline segments resulting
* in a single read-only component. The ``old'' segments are discarded when no scanner is reading
* them.
*/
@InterfaceAudience.Private
@ -62,7 +62,7 @@ public class CompactingMemStore extends AbstractMemStore {
// Default fraction of in-memory-flush size w.r.t. flush-to-disk size
public static final String IN_MEMORY_FLUSH_THRESHOLD_FACTOR_KEY =
"hbase.memstore.inmemoryflush.threshold.factor";
private static final double IN_MEMORY_FLUSH_THRESHOLD_FACTOR_DEFAULT = 0.1;
private static final double IN_MEMORY_FLUSH_THRESHOLD_FACTOR_DEFAULT = 0.014;
private static final Logger LOG = LoggerFactory.getLogger(CompactingMemStore.class);
private HStore store;

2
hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/MemStoreCompactionStrategy.java
View File

@ -42,7 +42,7 @@ public abstract class MemStoreCompactionStrategy {
// The upper bound for the number of segments we store in the pipeline prior to merging.
public static final String COMPACTING_MEMSTORE_THRESHOLD_KEY =
"hbase.hregion.compacting.pipeline.segments.limit";
public static final int COMPACTING_MEMSTORE_THRESHOLD_DEFAULT = 4;
public static final int COMPACTING_MEMSTORE_THRESHOLD_DEFAULT = 2;
/**
* Types of actions to be done on the pipeline upon MemStoreCompaction invocation.

4
hbase-server/src/test/java/org/apache/hadoop/hbase/regionserver/TestCompactingToCellFlatMapMemStore.java
View File

@ -848,8 +848,10 @@ public class TestCompactingToCellFlatMapMemStore extends TestCompactingMemStore
// set memstore to flat into CellChunkMap
MemoryCompactionPolicy compactionType = MemoryCompactionPolicy.BASIC;
memstore.getConfiguration().setInt(MemStoreCompactionStrategy
.COMPACTING_MEMSTORE_THRESHOLD_KEY, 4);
memstore.getConfiguration().set(CompactingMemStore.COMPACTING_MEMSTORE_TYPE_KEY,
String.valueOf(compactionType));
String.valueOf(compactionType));
((MyCompactingMemStore) memstore).initiateType(compactionType, memstore.getConfiguration());
((CompactingMemStore) memstore).setIndexType(CompactingMemStore.IndexType.CHUNK_MAP);

7
src/main/asciidoc/_chapters/inmemory_compaction.adoc
View File

@ -52,7 +52,7 @@ In this section we describe how to enable Accordion and the available configurat
To enable in-memory compactions, set the _IN_MEMORY_COMPACTION_ attribute
on per column family where you want the behavior. The _IN_MEMORY_COMPACTION_
attribute can have one of three values.
attribute can have one of four values.
* _NONE_: No in-memory compaction.
* _BASIC_: Basic policy enables flushing and keeps a pipeline of flushes until we trip the pipeline maximum threshold and then we flush to disk. No in-memory compaction but can help throughput as data is moved from the profligate, native ConcurrentSkipListMap data-type to more compact (and efficient) data types.
@ -94,11 +94,12 @@ to set the in-memory attribute to _NONE_).
When an in-memory flush happens is calculated by dividing the configured region flush size (Set in the table descriptor
or read from _hbase.hregion.memstore.flush.size_) by the number of column families and then multiplying by
_hbase.memstore.inmemoryflush.threshold.factor_ (default is 0.1).
_hbase.memstore.inmemoryflush.threshold.factor_ (default is 0.014 - a little less than the size
of a chunk).
The number of flushes carried by the pipeline is monitored so as to fit within the bounds of memstore sizing
but you can also set a maximum on the number of flushes total by setting
_hbase.hregion.compacting.pipeline.segments.limit_. Default is 4.
_hbase.hregion.compacting.pipeline.segments.limit_. Default is 2.
When a column family Store is created, it says what memstore type is in effect. As of this writing
there is the old-school _DefaultMemStore_ which fills a _ConcurrentSkipListMap_ and then flushes