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

HBASE-5356 region_mover.rb can hang if table region it belongs to is deleted

Browse Source 
git-svn-id: https://svn.apache.org/repos/asf/hbase/trunk@1565742 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
jxiang 2014-02-07 17:57:44 +00:00
parent d30249534b
commit 147a2304d1
1 changed files with 16 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
bin/region_mover.rb
View File

@ -82,8 +82,7 @@ end
# Returns true if passed region is still on 'original' when we look at .META. # Returns true if passed region is still on 'original' when we look at .META.
def isSameServer(admin, r, original) def isSameServer(admin, r, original)
server = getServerNameForRegion(admin, r) server = getServerNameForRegion(admin, r)
return false unless server return false unless server and original
return true unless original
return server == original return server == original
end end
@ -97,6 +96,7 @@ end
# Get servername that is up in .META.; this is hostname + port + startcode comma-delimited. # Get servername that is up in .META.; this is hostname + port + startcode comma-delimited.
# Can return nil # Can return nil
def getServerNameForRegion(admin, r) def getServerNameForRegion(admin, r)
return nil unless admin.isTableEnabled(r.getTableName)
if r.isMetaRegion() if r.isMetaRegion()
# Hack # Hack
zkw = org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher.new(admin.getConfiguration(), "region_mover", nil) zkw = org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher.new(admin.getConfiguration(), "region_mover", nil)
@ -120,6 +120,7 @@ def getServerNameForRegion(admin, r)
g.addColumn(HConstants::CATALOG_FAMILY, HConstants::SERVER_QUALIFIER) g.addColumn(HConstants::CATALOG_FAMILY, HConstants::SERVER_QUALIFIER)
g.addColumn(HConstants::CATALOG_FAMILY, HConstants::STARTCODE_QUALIFIER) g.addColumn(HConstants::CATALOG_FAMILY, HConstants::STARTCODE_QUALIFIER)
result = table.get(g) result = table.get(g)
return nil unless result
server = result.getValue(HConstants::CATALOG_FAMILY, HConstants::SERVER_QUALIFIER) server = result.getValue(HConstants::CATALOG_FAMILY, HConstants::SERVER_QUALIFIER)
startcode = result.getValue(HConstants::CATALOG_FAMILY, HConstants::STARTCODE_QUALIFIER) startcode = result.getValue(HConstants::CATALOG_FAMILY, HConstants::STARTCODE_QUALIFIER)
return nil unless server return nil unless server
@ -133,8 +134,15 @@ def isSuccessfulScan(admin, r)
scan.setBatch(1) scan.setBatch(1)
scan.setCaching(1) scan.setCaching(1)
scan.setFilter(FirstKeyOnlyFilter.new()) scan.setFilter(FirstKeyOnlyFilter.new())
table = getTable(admin.getConfiguration(), r.getTableName()) begin
scanner = table.getScanner(scan) table = getTable(admin.getConfiguration(), r.getTableName())
scanner = table.getScanner(scan)
rescue org.apache.hadoop.hbase.TableNotFoundException,
org.apache.hadoop.hbase.TableNotEnabledException => e
$LOG.warn("Region " + r.getEncodedName() + " belongs to recently " +
"deleted/disabled table. Skipping... " + e.message)
return
end
begin begin
results = scanner.next() results = scanner.next()
# We might scan into next region, this might be an empty table. # We might scan into next region, this might be an empty table.
@ -164,7 +172,8 @@ def move(admin, r, newServer, original)
count = count + 1 count = count + 1
begin begin
admin.move(Bytes.toBytes(r.getEncodedName()), Bytes.toBytes(newServer)) admin.move(Bytes.toBytes(r.getEncodedName()), Bytes.toBytes(newServer))
rescue java.lang.reflect.UndeclaredThrowableException => e rescue java.lang.reflect.UndeclaredThrowableException,
org.apache.hadoop.hbase.UnknownRegionException => e
$LOG.info("Exception moving " + r.getEncodedName() + $LOG.info("Exception moving " + r.getEncodedName() +
"; split/moved? Continuing: " + e) "; split/moved? Continuing: " + e)
return return
@ -332,6 +341,8 @@ def unloadRegions(options, hostname)
movedRegions = java.util.ArrayList.new() movedRegions = java.util.ArrayList.new()
while true while true
rs = getRegions(config, servername) rs = getRegions(config, servername)
# Remove those already tried to move
rs.removeAll(movedRegions)
break if rs.length == 0 break if rs.length == 0
count = 0 count = 0
$LOG.info("Moving " + rs.length.to_s + " region(s) from " + servername + $LOG.info("Moving " + rs.length.to_s + " region(s) from " + servername +