[
https://issues.apache.org/jira/browse/HBASE-20697?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16545899#comment-16545899
]
huaxiang sun commented on HBASE-20697:
--------------------------------------
[~stack] The fix is generic, getAllRegionLocations is not caching all regions'
locations, instead, it only caches the first entry. With the fix, the case for
region replicas is also taken care of. I think we need to backport this to 1.2
and 1.3 as well.
> Can't cache All region locations of the specify table by calling
> table.getRegionLocator().getAllRegionLocations()
> -----------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-20697
> URL: https://issues.apache.org/jira/browse/HBASE-20697
> Project: HBase
> Issue Type: Bug
> Affects Versions: 1.3.1, 1.2.6, 2.0.1
> Reporter: zhaoyuan
> Assignee: zhaoyuan
> Priority: Major
> Fix For: 3.0.0, 1.5.0, 1.4.6, 2.0.2, 2.2.0, 2.1.1
>
> Attachments: HBASE-20697.branch-1.2.001.patch,
> HBASE-20697.branch-1.2.002.patch, HBASE-20697.branch-1.2.003.patch,
> HBASE-20697.branch-1.2.004.patch, HBASE-20697.branch-1.addendum.patch,
> HBASE-20697.master.001.patch, HBASE-20697.master.002.patch,
> HBASE-20697.master.002.patch, HBASE-20697.master.003.patch
>
>
> When we upgrade and restart a new version application which will read and
> write to HBase, we will get some operation timeout. The time out is expected
> because when the application restarts，It will not hold any region locations
> cache and do communication with zk and meta regionserver to get region
> locations.
> We want to avoid these timeouts so we do warmup work and as far as I am
> concerned,the method table.getRegionLocator().getAllRegionLocations() will
> fetch all region locations and cache them. However, it didn't work good.
> There are still a lot of time outs,so it confused me.
> I dig into the source code and find something below
> {code:java}
> // code placeholder
> public List<HRegionLocation> getAllRegionLocations() throws IOException {
> TableName tableName = getName();
> NavigableMap<HRegionInfo, ServerName> locations =
> MetaScanner.allTableRegions(this.connection, tableName);
> ArrayList<HRegionLocation> regions = new ArrayList<>(locations.size());
> for (Entry<HRegionInfo, ServerName> entry : locations.entrySet()) {
> regions.add(new HRegionLocation(entry.getKey(), entry.getValue()));
> }
> if (regions.size() > 0) {
> connection.cacheLocation(tableName, new RegionLocations(regions));
> }
> return regions;
> }
> In MetaCache
> public void cacheLocation(final TableName tableName, final RegionLocations
> locations) {
> byte [] startKey =
> locations.getRegionLocation().getRegionInfo().getStartKey();
> ConcurrentMap<byte[], RegionLocations> tableLocations =
> getTableLocations(tableName);
> RegionLocations oldLocation = tableLocations.putIfAbsent(startKey,
> locations);
> boolean isNewCacheEntry = (oldLocation == null);
> if (isNewCacheEntry) {
> if (LOG.isTraceEnabled()) {
> LOG.trace("Cached location: " + locations);
> }
> addToCachedServers(locations);
> return;
> }
> {code}
> It will collect all regions into one RegionLocations object and only cache
> the first not null region location and then when we put or get to hbase, we
> do getCacheLocation()
> {code:java}
> // code placeholder
> public RegionLocations getCachedLocation(final TableName tableName, final
> byte [] row) {
> ConcurrentNavigableMap<byte[], RegionLocations> tableLocations =
> getTableLocations(tableName);
> Entry<byte[], RegionLocations> e = tableLocations.floorEntry(row);
> if (e == null) {
> if (metrics!= null) metrics.incrMetaCacheMiss();
> return null;
> }
> RegionLocations possibleRegion = e.getValue();
> // make sure that the end key is greater than the row we're looking
> // for, otherwise the row actually belongs in the next region, not
> // this one. the exception case is when the endkey is
> // HConstants.EMPTY_END_ROW, signifying that the region we're
> // checking is actually the last region in the table.
> byte[] endKey =
> possibleRegion.getRegionLocation().getRegionInfo().getEndKey();
> if (Bytes.equals(endKey, HConstants.EMPTY_END_ROW) ||
> getRowComparator(tableName).compareRows(
> endKey, 0, endKey.length, row, 0, row.length) > 0) {
> if (metrics != null) metrics.incrMetaCacheHit();
> return possibleRegion;
> }
> // Passed all the way through, so we got nothing - complete cache miss
> if (metrics != null) metrics.incrMetaCacheMiss();
> return null;
> }
> {code}
> It will choose the first location to be possibleRegion and possibly it will
> miss match.
> So did I forget something or may be wrong somewhere? If this is indeed a bug
> I think it can be fixed not very hard.
> Hope commiters and PMC review this !
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
