@Override
public void processCommit(CommitUpdateCommand cmd) throws IOException {
if (zkEnabled) {
zkCheck();
}
if (vinfo != null) {
vinfo.lockForUpdate();
}
try {
if (ulog == null
|| ulog.getState() == UpdateLog.State.ACTIVE
|| (cmd.getFlags() & UpdateCommand.REPLAY) != 0) {
super.processCommit(cmd);
} else {
log.info(
"Ignoring commit while not ACTIVE - state: "
+ ulog.getState()
+ " replay:"
+ (cmd.getFlags() & UpdateCommand.REPLAY));
}
} finally {
if (vinfo != null) {
vinfo.unlockForUpdate();
}
}
// TODO: we should consider this? commit everyone in the current collection
if (zkEnabled) {
ModifiableSolrParams params = new ModifiableSolrParams(filterParams(req.getParams()));
if (!req.getParams().getBool(COMMIT_END_POINT, false)) {
params.set(COMMIT_END_POINT, true);
String nodeName =
req.getCore().getCoreDescriptor().getCoreContainer().getZkController().getNodeName();
String shardZkNodeName = nodeName + "_" + req.getCore().getName();
List<Node> nodes =
getCollectionUrls(
req,
req.getCore().getCoreDescriptor().getCloudDescriptor().getCollectionName(),
shardZkNodeName);
if (nodes != null) {
cmdDistrib.distribCommit(cmd, nodes, params);
finish();
}
}
}
}
private boolean versionDelete(DeleteUpdateCommand cmd) throws IOException {
BytesRef idBytes = cmd.getIndexedId();
if (vinfo == null || idBytes == null) {
super.processDelete(cmd);
return false;
}
// This is only the hash for the bucket, and must be based only on the uniqueKey (i.e. do not
// use a pluggable hash here)
int bucketHash = Hash.murmurhash3_x86_32(idBytes.bytes, idBytes.offset, idBytes.length, 0);
// at this point, there is an update we need to try and apply.
// we may or may not be the leader.
// Find the version
long versionOnUpdate = cmd.getVersion();
if (versionOnUpdate == 0) {
String versionOnUpdateS = req.getParams().get(VERSION_FIELD);
versionOnUpdate = versionOnUpdateS == null ? 0 : Long.parseLong(versionOnUpdateS);
}
long signedVersionOnUpdate = versionOnUpdate;
versionOnUpdate = Math.abs(versionOnUpdate); // normalize to positive version
boolean isReplay = (cmd.getFlags() & UpdateCommand.REPLAY) != 0;
boolean leaderLogic = isLeader && !isReplay;
if (!leaderLogic && versionOnUpdate == 0) {
throw new SolrException(ErrorCode.BAD_REQUEST, "missing _version_ on update from leader");
}
VersionBucket bucket = vinfo.bucket(bucketHash);
vinfo.lockForUpdate();
try {
synchronized (bucket) {
if (versionsStored) {
long bucketVersion = bucket.highest;
if (leaderLogic) {
if (signedVersionOnUpdate != 0) {
Long lastVersion = vinfo.lookupVersion(cmd.getIndexedId());
long foundVersion = lastVersion == null ? -1 : lastVersion;
if ((signedVersionOnUpdate == foundVersion)
|| (signedVersionOnUpdate < 0 && foundVersion < 0)
|| (signedVersionOnUpdate == 1 && foundVersion > 0)) {
// we're ok if versions match, or if both are negative (all missing docs are equal),
// or if cmd
// specified it must exist (versionOnUpdate==1) and it does.
} else {
throw new SolrException(
ErrorCode.CONFLICT,
"version conflict for "
+ cmd.getId()
+ " expected="
+ signedVersionOnUpdate
+ " actual="
+ foundVersion);
}
}
long version = vinfo.getNewClock();
cmd.setVersion(-version);
bucket.updateHighest(version);
} else {
cmd.setVersion(-versionOnUpdate);
if (ulog.getState() != UpdateLog.State.ACTIVE
&& (cmd.getFlags() & UpdateCommand.REPLAY) == 0) {
// we're not in an active state, and this update isn't from a replay, so buffer it.
cmd.setFlags(cmd.getFlags() | UpdateCommand.BUFFERING);
ulog.delete(cmd);
return true;
}
// if we aren't the leader, then we need to check that updates were not re-ordered
if (bucketVersion != 0 && bucketVersion < versionOnUpdate) {
// we're OK... this update has a version higher than anything we've seen
// in this bucket so far, so we know that no reordering has yet occured.
bucket.updateHighest(versionOnUpdate);
} else {
// there have been updates higher than the current update. we need to check
// the specific version for this id.
Long lastVersion = vinfo.lookupVersion(cmd.getIndexedId());
if (lastVersion != null && Math.abs(lastVersion) >= versionOnUpdate) {
// This update is a repeat, or was reordered. We need to drop this update.
return true;
}
}
}
}
doLocalDelete(cmd);
return false;
} // end synchronized (bucket)
} finally {
vinfo.unlockForUpdate();
}
}
public void doDeleteByQuery(DeleteUpdateCommand cmd) throws IOException {
// even in non zk mode, tests simulate updates from a leader
if (!zkEnabled) {
isLeader = getNonZkLeaderAssumption(req);
} else {
zkCheck();
}
// NONE: we are the first to receive this deleteByQuery
// - it must be forwarded to the leader of every shard
// TO: we are a leader receiving a forwarded deleteByQuery... we must:
// - block all updates (use VersionInfo)
// - flush *all* updates going to our replicas
// - forward the DBQ to our replicas and wait for the response
// - log + execute the local DBQ
// FROM: we are a replica receiving a DBQ from our leader
// - log + execute the local DBQ
DistribPhase phase = DistribPhase.parseParam(req.getParams().get(DISTRIB_UPDATE_PARAM));
if (zkEnabled && DistribPhase.NONE == phase) {
boolean leaderForAnyShard = false; // start off by assuming we are not a leader for any shard
Map<String, Slice> slices = zkController.getClusterState().getSlices(collection);
if (slices == null) {
throw new SolrException(
ErrorCode.BAD_REQUEST,
"Cannot find collection:"
+ collection
+ " in "
+ zkController.getClusterState().getCollections());
}
ModifiableSolrParams params = new ModifiableSolrParams(filterParams(req.getParams()));
params.set(DISTRIB_UPDATE_PARAM, DistribPhase.TOLEADER.toString());
List<Node> leaders = new ArrayList<Node>(slices.size());
for (Map.Entry<String, Slice> sliceEntry : slices.entrySet()) {
String sliceName = sliceEntry.getKey();
ZkNodeProps leaderProps;
try {
leaderProps = zkController.getZkStateReader().getLeaderProps(collection, sliceName);
} catch (InterruptedException e) {
throw new SolrException(
ErrorCode.SERVICE_UNAVAILABLE, "Exception finding leader for shard " + sliceName, e);
}
// TODO: What if leaders changed in the meantime?
// should we send out slice-at-a-time and if a node returns "hey, I'm not a leader" (or we
// get an error because it went down) then look up the new leader?
// Am I the leader for this slice?
ZkCoreNodeProps coreLeaderProps = new ZkCoreNodeProps(leaderProps);
String leaderNodeName = coreLeaderProps.getCoreNodeName();
String coreName = req.getCore().getName();
String coreNodeName = zkController.getNodeName() + "_" + coreName;
isLeader = coreNodeName.equals(leaderNodeName);
if (isLeader) {
// don't forward to ourself
leaderForAnyShard = true;
} else {
leaders.add(new StdNode(coreLeaderProps));
}
}
params.remove("commit"); // this will be distributed from the local commit
cmdDistrib.distribDelete(cmd, leaders, params);
if (!leaderForAnyShard) {
return;
}
// change the phase to TOLEADER so we look up and forward to our own replicas (if any)
phase = DistribPhase.TOLEADER;
}
List<Node> replicas = null;
if (zkEnabled && DistribPhase.TOLEADER == phase) {
// This core should be a leader
isLeader = true;
replicas = setupRequest();
} else if (DistribPhase.FROMLEADER == phase) {
isLeader = false;
}
if (vinfo == null) {
super.processDelete(cmd);
return;
}
// at this point, there is an update we need to try and apply.
// we may or may not be the leader.
// Find the version
long versionOnUpdate = cmd.getVersion();
if (versionOnUpdate == 0) {
String versionOnUpdateS = req.getParams().get(VERSION_FIELD);
versionOnUpdate = versionOnUpdateS == null ? 0 : Long.parseLong(versionOnUpdateS);
}
versionOnUpdate = Math.abs(versionOnUpdate); // normalize to positive version
boolean isReplay = (cmd.getFlags() & UpdateCommand.REPLAY) != 0;
boolean leaderLogic = isLeader && !isReplay;
if (!leaderLogic && versionOnUpdate == 0) {
throw new SolrException(ErrorCode.BAD_REQUEST, "missing _version_ on update from leader");
}
vinfo.blockUpdates();
try {
if (versionsStored) {
if (leaderLogic) {
long version = vinfo.getNewClock();
cmd.setVersion(-version);
// TODO update versions in all buckets
doLocalDelete(cmd);
} else {
cmd.setVersion(-versionOnUpdate);
if (ulog.getState() != UpdateLog.State.ACTIVE
&& (cmd.getFlags() & UpdateCommand.REPLAY) == 0) {
// we're not in an active state, and this update isn't from a replay, so buffer it.
cmd.setFlags(cmd.getFlags() | UpdateCommand.BUFFERING);
ulog.deleteByQuery(cmd);
return;
}
doLocalDelete(cmd);
}
}
// since we don't know which documents were deleted, the easiest thing to do is to invalidate
// all real-time caches (i.e. UpdateLog) which involves also getting a new version of the
// IndexReader
// (so cache misses will see up-to-date data)
} finally {
vinfo.unblockUpdates();
}
// forward to all replicas
if (leaderLogic && replicas != null) {
ModifiableSolrParams params = new ModifiableSolrParams(filterParams(req.getParams()));
params.set(VERSION_FIELD, Long.toString(cmd.getVersion()));
params.set(DISTRIB_UPDATE_PARAM, DistribPhase.FROMLEADER.toString());
params.set(
"update.from",
ZkCoreNodeProps.getCoreUrl(zkController.getBaseUrl(), req.getCore().getName()));
cmdDistrib.distribDelete(cmd, replicas, params);
cmdDistrib.finish();
}
if (returnVersions && rsp != null) {
if (deleteByQueryResponse == null) {
deleteByQueryResponse = new NamedList<String>();
rsp.add("deleteByQuery", deleteByQueryResponse);
}
deleteByQueryResponse.add(cmd.getQuery(), cmd.getVersion());
}
}
/**
* @return whether or not to drop this cmd
* @throws IOException If there is a low-level I/O error.
*/
private boolean versionAdd(AddUpdateCommand cmd) throws IOException {
BytesRef idBytes = cmd.getIndexedId();
if (vinfo == null || idBytes == null) {
super.processAdd(cmd);
return false;
}
// This is only the hash for the bucket, and must be based only on the uniqueKey (i.e. do not
// use a pluggable hash here)
int bucketHash = Hash.murmurhash3_x86_32(idBytes.bytes, idBytes.offset, idBytes.length, 0);
// at this point, there is an update we need to try and apply.
// we may or may not be the leader.
// Find any existing version in the document
// TODO: don't reuse update commands any more!
long versionOnUpdate = cmd.getVersion();
if (versionOnUpdate == 0) {
SolrInputField versionField = cmd.getSolrInputDocument().getField(VersionInfo.VERSION_FIELD);
if (versionField != null) {
Object o = versionField.getValue();
versionOnUpdate =
o instanceof Number ? ((Number) o).longValue() : Long.parseLong(o.toString());
} else {
// Find the version
String versionOnUpdateS = req.getParams().get(VERSION_FIELD);
versionOnUpdate = versionOnUpdateS == null ? 0 : Long.parseLong(versionOnUpdateS);
}
}
boolean isReplay = (cmd.getFlags() & UpdateCommand.REPLAY) != 0;
boolean leaderLogic = isLeader && !isReplay;
VersionBucket bucket = vinfo.bucket(bucketHash);
vinfo.lockForUpdate();
try {
synchronized (bucket) {
// we obtain the version when synchronized and then do the add so we can ensure that
// if version1 < version2 then version1 is actually added before version2.
// even if we don't store the version field, synchronizing on the bucket
// will enable us to know what version happened first, and thus enable
// realtime-get to work reliably.
// TODO: if versions aren't stored, do we need to set on the cmd anyway for some reason?
// there may be other reasons in the future for a version on the commands
boolean checkDeleteByQueries = false;
if (versionsStored) {
long bucketVersion = bucket.highest;
if (leaderLogic) {
boolean updated = getUpdatedDocument(cmd, versionOnUpdate);
if (versionOnUpdate != 0) {
Long lastVersion = vinfo.lookupVersion(cmd.getIndexedId());
long foundVersion = lastVersion == null ? -1 : lastVersion;
if (versionOnUpdate == foundVersion
|| (versionOnUpdate < 0 && foundVersion < 0)
|| (versionOnUpdate == 1 && foundVersion > 0)) {
// we're ok if versions match, or if both are negative (all missing docs are equal),
// or if cmd
// specified it must exist (versionOnUpdate==1) and it does.
} else {
throw new SolrException(
ErrorCode.CONFLICT,
"version conflict for "
+ cmd.getPrintableId()
+ " expected="
+ versionOnUpdate
+ " actual="
+ foundVersion);
}
}
long version = vinfo.getNewClock();
cmd.setVersion(version);
cmd.getSolrInputDocument().setField(VersionInfo.VERSION_FIELD, version);
bucket.updateHighest(version);
} else {
// The leader forwarded us this update.
cmd.setVersion(versionOnUpdate);
if (ulog.getState() != UpdateLog.State.ACTIVE
&& (cmd.getFlags() & UpdateCommand.REPLAY) == 0) {
// we're not in an active state, and this update isn't from a replay, so buffer it.
cmd.setFlags(cmd.getFlags() | UpdateCommand.BUFFERING);
ulog.add(cmd);
return true;
}
// if we aren't the leader, then we need to check that updates were not re-ordered
if (bucketVersion != 0 && bucketVersion < versionOnUpdate) {
// we're OK... this update has a version higher than anything we've seen
// in this bucket so far, so we know that no reordering has yet occurred.
bucket.updateHighest(versionOnUpdate);
} else {
// there have been updates higher than the current update. we need to check
// the specific version for this id.
Long lastVersion = vinfo.lookupVersion(cmd.getIndexedId());
if (lastVersion != null && Math.abs(lastVersion) >= versionOnUpdate) {
// This update is a repeat, or was reordered. We need to drop this update.
return true;
}
// also need to re-apply newer deleteByQuery commands
checkDeleteByQueries = true;
}
}
}
boolean willDistrib = isLeader && nodes != null && nodes.size() > 0;
SolrInputDocument clonedDoc = null;
if (willDistrib) {
clonedDoc = cmd.solrDoc.deepCopy();
}
// TODO: possibly set checkDeleteByQueries as a flag on the command?
doLocalAdd(cmd);
if (willDistrib) {
cmd.solrDoc = clonedDoc;
}
} // end synchronized (bucket)
} finally {
vinfo.unlockForUpdate();
}
return false;
}