diff --git a/CHANGES.txt b/CHANGES.txt index 87c65defc3f..38f159e8146 100644 --- a/CHANGES.txt +++ b/CHANGES.txt @@ -133,6 +133,9 @@ Release 0.20.0 - Unreleased (Clint Morgan via Stack) HBASE-1431 NPE in HTable.checkAndSave when row doesn't exist (Guilherme Mauro Germoglio Barbosa via Andrew Purtell) + HBASE-1421 Processing a regionserver message -- OPEN, CLOSE, SPLIT, etc. -- + and if we're carrying more than one message in payload, if + exception, all messages that follow are dropped on floor IMPROVEMENTS HBASE-1089 Add count of regions on filesystem to master UI; add percentage diff --git a/src/java/org/apache/hadoop/hbase/master/ProcessRegionOpen.java b/src/java/org/apache/hadoop/hbase/master/ProcessRegionOpen.java index 7ab2b7b6379..5ebca6281fb 100644 --- a/src/java/org/apache/hadoop/hbase/master/ProcessRegionOpen.java +++ b/src/java/org/apache/hadoop/hbase/master/ProcessRegionOpen.java @@ -40,11 +40,9 @@ class ProcessRegionOpen extends ProcessRegionStatusChange { * @param master * @param info * @param regionInfo - * @throws IOException */ public ProcessRegionOpen(HMaster master, HServerInfo info, - HRegionInfo regionInfo) - throws IOException { + HRegionInfo regionInfo) { super(master, regionInfo); if (info == null) { throw new NullPointerException("HServerInfo cannot be null; " + diff --git a/src/java/org/apache/hadoop/hbase/master/RegionManager.java b/src/java/org/apache/hadoop/hbase/master/RegionManager.java index e7f15111edd..669689333f2 100644 --- a/src/java/org/apache/hadoop/hbase/master/RegionManager.java +++ b/src/java/org/apache/hadoop/hbase/master/RegionManager.java @@ -1230,9 +1230,8 @@ class RegionManager implements HConstants { */ synchronized void setPendingOpen(final String serverName) { if (!this.unassigned) { - throw new IllegalStateException( - "Cannot assign a region that is not currently unassigned. State: " + - toString()); + LOG.warn("Cannot assign a region that is not currently unassigned. " + + "FIX!! State: " + toString()); } this.unassigned = false; this.pendingOpen = true; @@ -1250,9 +1249,8 @@ class RegionManager implements HConstants { synchronized void setOpen() { if (!pendingOpen) { - throw new IllegalStateException( - "Cannot set a region as open if it has not been pending. State: " + - toString()); + LOG.warn("Cannot set a region as open if it has not been pending. " + + "FIX!! State: " + toString()); } this.unassigned = false; this.pendingOpen = false; @@ -1284,9 +1282,8 @@ class RegionManager implements HConstants { synchronized void setPendingClose() { if (!closing) { - throw new IllegalStateException( - "Cannot set a region as pending close if it has not been closing. " + - "State: " + toString()); + LOG.warn("Cannot set a region as pending close if it has not been " + + "closing. FIX!! State: " + toString()); } this.unassigned = false; this.pendingOpen = false; @@ -1353,4 +1350,4 @@ class RegionManager implements HConstants { return Bytes.compareTo(getRegionName(), o.getRegionName()); } } -} +} \ No newline at end of file diff --git a/src/java/org/apache/hadoop/hbase/master/RegionServerOperation.java b/src/java/org/apache/hadoop/hbase/master/RegionServerOperation.java index e11da11430b..169c0321bc3 100644 --- a/src/java/org/apache/hadoop/hbase/master/RegionServerOperation.java +++ b/src/java/org/apache/hadoop/hbase/master/RegionServerOperation.java @@ -42,7 +42,7 @@ abstract class RegionServerOperation implements Delayed, HConstants { // DelayQueue we're inserted in on lease expiration. this.expire = System.currentTimeMillis() + this.master.leaseTimeout / 2; } - + public long getDelay(TimeUnit unit) { return unit.convert(this.expire - System.currentTimeMillis(), TimeUnit.MILLISECONDS); diff --git a/src/java/org/apache/hadoop/hbase/master/ServerManager.java b/src/java/org/apache/hadoop/hbase/master/ServerManager.java index b9f3cffce59..921d1d00d9e 100644 --- a/src/java/org/apache/hadoop/hbase/master/ServerManager.java +++ b/src/java/org/apache/hadoop/hbase/master/ServerManager.java @@ -388,26 +388,30 @@ class ServerManager implements HConstants { return processMsgs(serverInfo, mostLoadedRegions, msgs); } - /** + /* * Process all the incoming messages from a server that's contacted us. - * * Note that we never need to update the server's load information because * that has already been done in regionServerReport. + * @param serverInfo + * @param mostLoadedRegions + * @param incomingMsgs + * @return */ private HMsg[] processMsgs(HServerInfo serverInfo, - HRegionInfo[] mostLoadedRegions, HMsg incomingMsgs[]) - throws IOException { + HRegionInfo[] mostLoadedRegions, HMsg incomingMsgs[]) { ArrayList returnMsgs = new ArrayList(); if (serverInfo.getServerAddress() == null) { throw new NullPointerException("Server address cannot be null; " + "hbase-958 debugging"); } // Get reports on what the RegionServer did. + // Be careful that in message processors we don't throw exceptions that + // break the switch below because then we might drop messages on the floor. int openingCount = 0; for (int i = 0; i < incomingMsgs.length; i++) { HRegionInfo region = incomingMsgs[i].getRegionInfo(); LOG.info("Received " + incomingMsgs[i] + " from " + - serverInfo.getServerName()); + serverInfo.getServerName() + "; " + i + " of " + incomingMsgs.length); switch (incomingMsgs[i].getType()) { case MSG_REPORT_PROCESS_OPEN: openingCount++; @@ -422,13 +426,11 @@ class ServerManager implements HConstants { break; case MSG_REPORT_SPLIT: - processSplitRegion(region, incomingMsgs[++i], incomingMsgs[++i], - returnMsgs); + processSplitRegion(region, incomingMsgs[++i], incomingMsgs[++i]); break; default: - throw new IOException( - "Impossible state during message processing. Instruction: " + + LOG.warn("Impossible state during message processing. Instruction: " + incomingMsgs[i].getType()); } } @@ -456,7 +458,7 @@ class ServerManager implements HConstants { return returnMsgs.toArray(new HMsg[returnMsgs.size()]); } - /** + /* * A region has split. * * @param region @@ -464,21 +466,16 @@ class ServerManager implements HConstants { * @param splitB * @param returnMsgs */ - private void processSplitRegion(HRegionInfo region, HMsg splitA, HMsg splitB, - ArrayList returnMsgs) { - + private void processSplitRegion(HRegionInfo region, HMsg splitA, HMsg splitB) { synchronized (master.regionManager) { // Cancel any actions pending for the affected region. // This prevents the master from sending a SPLIT message if the table // has already split by the region server. master.regionManager.endActions(region.getRegionName()); - HRegionInfo newRegionA = splitA.getRegionInfo(); master.regionManager.setUnassigned(newRegionA, false); - HRegionInfo newRegionB = splitB.getRegionInfo(); master.regionManager.setUnassigned(newRegionB, false); - if (region.isMetaTable()) { // A meta region has split. master.regionManager.offlineMetaRegion(region.getStartKey()); @@ -487,10 +484,14 @@ class ServerManager implements HConstants { } } - /** Region server is reporting that a region is now opened */ + /* + * Region server is reporting that a region is now opened + * @param serverInfo + * @param region + * @param returnMsgs + */ private void processRegionOpen(HServerInfo serverInfo, - HRegionInfo region, ArrayList returnMsgs) - throws IOException { + HRegionInfo region, ArrayList returnMsgs) { boolean duplicateAssignment = false; synchronized (master.regionManager) { if (!master.regionManager.isUnassigned(region) && @@ -549,19 +550,30 @@ class ServerManager implements HConstants { // Note that the table has been assigned and is waiting for the // meta table to be updated. master.regionManager.setOpen(region.getRegionNameAsString()); - // Queue up an update to note the region location. - try { - master.toDoQueue.put( - new ProcessRegionOpen(master, serverInfo, region)); - } catch (InterruptedException e) { - throw new RuntimeException( - "Putting into toDoQueue was interrupted.", e); + // Queue up an update to note the region location. Do inside + // a retry loop in case interrupted. + boolean succeeded = false; + for (int i = 0; i < 10; i++) { + try { + master.toDoQueue. + put(new ProcessRegionOpen(master, serverInfo, region)); + succeeded = true; + } catch (InterruptedException e) { + LOG.warn("Putting into toDoQueue was interrupted.", e); + } } - } + if (!succeeded) { + LOG.warn("FAILED ADDING OPEN TO TODO QUEUE: " + serverInfo); + } + } } } } - + + /* + * @param region + * @throws Exception + */ private void processRegionClose(HRegionInfo region) { synchronized (master.regionManager) { if (region.isRootRegion()) {