Skip to content

Commit

Permalink
ZOOKEEPER-2678: Discovery and Sync can take a very long time on large…
Browse files Browse the repository at this point in the history 
… DB (3.4)

This patch addresses recovery time when a leader is lost on a large DB.

It does this by not clearing the DB before leader election begins, and by avoiding taking a snapshot as part of the SYNC phase, specifically for a DIFF sync. It does this by buffering the proposals and commits just like the code currently does for proposals/commits sent after the NEWLEADER and before the UPTODATE messages.

If a SNAP is sent we cannot avoid writing out the full snapshot because there is no other way to make sure the disk DB is in sync with what is in memory.  So any edits to the edit log before a background snapshot happened could possibly be applied on top of an incorrect snapshot.

This same optimization should work for TRUNC too, but I opted not to do it for TRUNC because TRUNC is rare and TRUNC by its very nature already forces the DB to be reread after the edit logs are modified.  So it would still not be fast.

In practice this makes it so instead of taking 5+ mins for the cluster to recover from losing a leader it now takes about 3 seconds.

I am happy to port this to 3.5. if it looks good.

Author: Robert (Bobby) Evans <evans@yahoo-inc.com>

Reviewers: Flavio Junqueira <fpj@apache.org>, Edward Ribeiro <edward.ribeiro@gmail.com>, Abraham Fine <afine@apache.org>, Michael Han <hanm@apache.org>

Closes #157 from revans2/ZOOKEEPER-2678 and squashes the following commits:

d079617 [Robert (Bobby) Evans] ZOOKEEPER-2678:  Improved test to verify snapshot times
dcbf325 [Robert (Bobby) Evans] Addressed review comments
f57c384 [Robert (Bobby) Evans] ZOOKEEPER-2678: Fixed another race
f705293 [Robert (Bobby) Evans] ZOOKEEPER-2678: Addressed review comments
5aa2562 [Robert (Bobby) Evans] ZOOKEEPER-2678: Discovery and Sync can take a very long time on large DBs
  • Loading branch information
@hanm
Robert (Bobby) Evans authored and hanm committed Feb 14, 2017
1 parent e8247ee commit 2de93fe
Show file tree
Hide file tree
Showing 4 changed files with 67 additions and 16 deletions.
18 changes: 16 additions & 2 deletions src/java/main/org/apache/zookeeper/server/ZooKeeperServer.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -489,7 +489,15 @@ public boolean isRunning() {
return state == State.RUNNING;
}

public synchronized void shutdown() {
public void shutdown() {
shutdown(false);
}

/**
* Shut down the server instance
* @param fullyShutDown true if another server using the same database will not replace this one in the same process
*/
public synchronized void shutdown(boolean fullyShutDown) {
if (!canShutdown()) {
LOG.debug("ZooKeeper server is not running, so not proceeding to shutdown!");
return;
Expand All @@ -507,9 +515,15 @@ public synchronized void shutdown() {
if (firstProcessor != null) {
firstProcessor.shutdown();
}
if (zkDb != null) {

if (fullyShutDown && zkDb != null) {
zkDb.clear();
}
// else there is no need to clear the database
// * When a new quorum is established we can still apply the diff
// on top of the same zkDb data
// * If we fetch a new snapshot from leader, the zkDb will be
// cleared anyway before loading the snapshot

unregisterJMX();
}
Expand Down
2 changes: 1 addition & 1 deletion src/java/main/org/apache/zookeeper/server/ZooKeeperServerMain.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -124,7 +124,7 @@ public void runFromConfig(ServerConfig config) throws IOException {

cnxnFactory.join();
if (zkServer.canShutdown()) {
zkServer.shutdown();
zkServer.shutdown(true);
}
} catch (InterruptedException e) {
// warn, but generally this is ok
Expand Down
34 changes: 22 additions & 12 deletions src/java/main/org/apache/zookeeper/server/quorum/Learner.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -321,13 +321,16 @@ protected void syncWithLeader(long newLeaderZxid) throws IOException, Interrupte
QuorumPacket ack = new QuorumPacket(Leader.ACK, 0, null, null);
QuorumPacket qp = new QuorumPacket();
long newEpoch = ZxidUtils.getEpochFromZxid(newLeaderZxid);

readPacket(qp);
// In the DIFF case we don't need to do a snapshot because the transactions will sync on top of any existing snapshot
// For SNAP and TRUNC the snapshot is needed to save that history
boolean snapshotNeeded = true;
readPacket(qp);
LinkedList<Long> packetsCommitted = new LinkedList<Long>();
LinkedList<PacketInFlight> packetsNotCommitted = new LinkedList<PacketInFlight>();
synchronized (zk) {
if (qp.getType() == Leader.DIFF) {
LOG.info("Getting a diff from the leader 0x" + Long.toHexString(qp.getZxid()));
LOG.info("Getting a diff from the leader 0x{}", Long.toHexString(qp.getZxid()));
snapshotNeeded = false;
}
else if (qp.getType() == Leader.SNAP) {
LOG.info("Getting a snapshot from leader");
Expand Down Expand Up @@ -364,10 +367,13 @@ else if (qp.getType() == Leader.SNAP) {

long lastQueued = 0;

// in V1.0 we take a snapshot when we get the NEWLEADER message, but in pre V1.0
// we take the snapshot at the UPDATE, since V1.0 also gets the UPDATE (after the NEWLEADER)
// in Zab V1.0 (ZK 3.4+) we might take a snapshot when we get the NEWLEADER message, but in pre V1.0
// we take the snapshot on the UPDATE message, since Zab V1.0 also gets the UPDATE (after the NEWLEADER)
// we need to make sure that we don't take the snapshot twice.
boolean snapshotTaken = false;
boolean isPreZAB1_0 = true;
//If we are not going to take the snapshot be sure the transactions are not applied in memory
// but written out to the transaction log
boolean writeToTxnLog = !snapshotNeeded;
// we are now going to start getting transactions to apply followed by an UPTODATE
outerLoop:
while (self.isRunning()) {
Expand All @@ -387,7 +393,7 @@ else if (qp.getType() == Leader.SNAP) {
packetsNotCommitted.add(pif);
break;
case Leader.COMMIT:
if (!snapshotTaken) {
if (!writeToTxnLog) {
pif = packetsNotCommitted.peekFirst();
if (pif.hdr.getZxid() != qp.getZxid()) {
LOG.warn("Committing " + qp.getZxid() + ", but next proposal is " + pif.hdr.getZxid());
Expand Down Expand Up @@ -415,7 +421,7 @@ else if (qp.getType() == Leader.SNAP) {
+ Long.toHexString(lastQueued + 1));
}
lastQueued = packet.hdr.getZxid();
if (!snapshotTaken) {
if (!writeToTxnLog) {
// Apply to db directly if we haven't taken the snapshot
zk.processTxn(packet.hdr, packet.rec);
} else {
Expand All @@ -424,13 +430,14 @@ else if (qp.getType() == Leader.SNAP) {
}
break;
case Leader.UPTODATE:
if (!snapshotTaken) { // true for the pre v1.0 case
if (isPreZAB1_0) {
zk.takeSnapshot();
self.setCurrentEpoch(newEpoch);
}
self.cnxnFactory.setZooKeeperServer(zk);
break outerLoop;
case Leader.NEWLEADER: // it will be NEWLEADER in v1.0
case Leader.NEWLEADER: // Getting NEWLEADER here instead of in discovery
// means this is Zab 1.0
// Create updatingEpoch file and remove it after current
// epoch is set. QuorumPeer.loadDataBase() uses this file to
// detect the case where the server was terminated after
Expand All @@ -441,13 +448,16 @@ else if (qp.getType() == Leader.SNAP) {
throw new IOException("Failed to create " +
updating.toString());
}
zk.takeSnapshot();
if (snapshotNeeded) {
zk.takeSnapshot();
}
self.setCurrentEpoch(newEpoch);
if (!updating.delete()) {
throw new IOException("Failed to delete " +
updating.toString());
}
snapshotTaken = true;
writeToTxnLog = true; //Anything after this needs to go to the transaction log, not applied directly in memory
isPreZAB1_0 = false;
writePacket(new QuorumPacket(Leader.ACK, newLeaderZxid, null, null), true);
break;
}
Expand Down
29 changes: 28 additions & 1 deletion src/java/test/org/apache/zookeeper/server/quorum/Zab1_0Test.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,9 @@
package org.apache.zookeeper.server.quorum;

import static org.junit.Assert.assertEquals;
import static org.mockito.Mockito.never;
import static org.mockito.Mockito.spy;
import static org.mockito.Mockito.verify;

import java.io.BufferedInputStream;
import java.io.BufferedReader;
Expand Down Expand Up @@ -645,6 +648,8 @@ public void converseWithFollower(InputArchive ia, OutputArchive oa,
tmpDir.mkdir();
File logDir = f.fzk.getTxnLogFactory().getDataDir().getParentFile();
File snapDir = f.fzk.getTxnLogFactory().getSnapDir().getParentFile();
//Spy on ZK so we can check if a snapshot happened or not.
f.zk = spy(f.zk);
try {
Assert.assertEquals(0, f.self.getAcceptedEpoch());
Assert.assertEquals(0, f.self.getCurrentEpoch());
Expand Down Expand Up @@ -687,6 +692,10 @@ public void converseWithFollower(InputArchive ia, OutputArchive oa,
oa.writeRecord(qp, null);
zkDb.serializeSnapshot(oa);
oa.writeString("BenWasHere", null);
Thread.sleep(10); //Give it some time to process the snap
//No Snapshot taken yet, the SNAP was applied in memory
verify(f.zk, never()).takeSnapshot();

qp.setType(Leader.NEWLEADER);
qp.setZxid(ZxidUtils.makeZxid(1, 0));
oa.writeRecord(qp, null);
Expand All @@ -697,7 +706,8 @@ public void converseWithFollower(InputArchive ia, OutputArchive oa,
Assert.assertEquals(ZxidUtils.makeZxid(1, 0), qp.getZxid());
Assert.assertEquals(1, f.self.getAcceptedEpoch());
Assert.assertEquals(1, f.self.getCurrentEpoch());

//Make sure that we did take the snapshot now
verify(f.zk).takeSnapshot();
Assert.assertEquals(firstZxid, f.fzk.getLastProcessedZxid());

// Make sure the data was recorded in the filesystem ok
Expand Down Expand Up @@ -773,6 +783,8 @@ public void converseWithFollower(InputArchive ia, OutputArchive oa,
tmpDir.mkdir();
File logDir = f.fzk.getTxnLogFactory().getDataDir().getParentFile();
File snapDir = f.fzk.getTxnLogFactory().getSnapDir().getParentFile();
//Spy on ZK so we can check if a snapshot happened or not.
f.zk = spy(f.zk);
try {
Assert.assertEquals(0, f.self.getAcceptedEpoch());
Assert.assertEquals(0, f.self.getCurrentEpoch());
Expand Down Expand Up @@ -839,13 +851,28 @@ public void converseWithFollower(InputArchive ia, OutputArchive oa,
Assert.assertEquals(1, f.self.getAcceptedEpoch());
Assert.assertEquals(1, f.self.getCurrentEpoch());

//Wait for the transactions to be written out. The thread that writes them out
// does not send anything back when it is done.
long start = System.currentTimeMillis();
while (createSessionZxid != f.fzk.getLastProcessedZxid() && (System.currentTimeMillis() - start) < 50) {
Thread.sleep(1);
}

Assert.assertEquals(createSessionZxid, f.fzk.getLastProcessedZxid());

// Make sure the data was recorded in the filesystem ok
ZKDatabase zkDb2 = new ZKDatabase(new FileTxnSnapLog(logDir, snapDir));
start = System.currentTimeMillis();
zkDb2.loadDataBase();
while (zkDb2.getSessionWithTimeOuts().isEmpty() && (System.currentTimeMillis() - start) < 50) {
Thread.sleep(1);
zkDb2.loadDataBase();
}
LOG.info("zkdb2 sessions:" + zkDb2.getSessions());
LOG.info("zkdb2 with timeouts:" + zkDb2.getSessionWithTimeOuts());
Assert.assertNotNull(zkDb2.getSessionWithTimeOuts().get(4L));
//Snapshot was never taken during very simple sync
verify(f.zk, never()).takeSnapshot();
} finally {
recursiveDelete(tmpDir);
}
Expand Down

0 comments on commit 2de93fe

Please sign in to comment.