use of org.apache.accumulo.core.metadata.TabletLocationState in project accumulo by apache.
the class TabletLocationStateTest method testGetState_Assigned.
@Test
public void testGetState_Assigned() throws Exception {
Set<TServerInstance> liveServers = new java.util.HashSet<>();
liveServers.add(future);
tls = new TabletLocationState(keyExtent, future, null, last, null, walogs, true);
assertEquals(TabletState.ASSIGNED, tls.getState(liveServers));
}
use of org.apache.accumulo.core.metadata.TabletLocationState in project accumulo by apache.
the class TabletLocationStateTest method testConstruction_FutureAndCurrent.
@Test
public void testConstruction_FutureAndCurrent() {
expect(keyExtent.toMetaRow()).andReturn(new Text("entry"));
replay(keyExtent);
var e = assertThrows(TabletLocationState.BadLocationStateException.class, () -> new TabletLocationState(keyExtent, future, current, last, null, walogs, true));
assertEquals(new Text("entry"), e.getEncodedEndRow());
}
use of org.apache.accumulo.core.metadata.TabletLocationState in project accumulo by apache.
the class TabletGroupWatcher method run.
@Override
public void run() {
int[] oldCounts = new int[TabletState.values().length];
EventCoordinator.Listener eventListener = this.manager.nextEvent.getListener();
WalStateManager wals = new WalStateManager(manager.getContext());
while (manager.stillManager()) {
// slow things down a little, otherwise we spam the logs when there are many wake-up events
sleepUninterruptibly(100, TimeUnit.MILLISECONDS);
int totalUnloaded = 0;
int unloaded = 0;
ClosableIterator<TabletLocationState> iter = null;
try {
Map<TableId, MergeStats> mergeStatsCache = new HashMap<>();
Map<TableId, MergeStats> currentMerges = new HashMap<>();
for (MergeInfo merge : manager.merges()) {
if (merge.getExtent() != null) {
currentMerges.put(merge.getExtent().tableId(), new MergeStats(merge));
}
}
// Get the current status for the current list of tservers
SortedMap<TServerInstance, TabletServerStatus> currentTServers = new TreeMap<>();
for (TServerInstance entry : manager.tserverSet.getCurrentServers()) {
currentTServers.put(entry, manager.tserverStatus.get(entry));
}
if (currentTServers.isEmpty()) {
eventListener.waitForEvents(Manager.TIME_TO_WAIT_BETWEEN_SCANS);
synchronized (this) {
lastScanServers = Collections.emptySortedSet();
}
continue;
}
TabletLists tLists = new TabletLists(manager, currentTServers);
ManagerState managerState = manager.getManagerState();
int[] counts = new int[TabletState.values().length];
stats.begin();
// Walk through the tablets in our store, and work tablets
// towards their goal
iter = store.iterator();
while (iter.hasNext()) {
TabletLocationState tls = iter.next();
if (tls == null) {
continue;
}
// ignore entries for tables that do not exist in zookeeper
if (manager.getTableManager().getTableState(tls.extent.tableId()) == null)
continue;
// Don't overwhelm the tablet servers with work
if (tLists.unassigned.size() + unloaded > Manager.MAX_TSERVER_WORK_CHUNK * currentTServers.size()) {
flushChanges(tLists, wals);
tLists.reset();
unloaded = 0;
eventListener.waitForEvents(Manager.TIME_TO_WAIT_BETWEEN_SCANS);
}
TableId tableId = tls.extent.tableId();
TableConfiguration tableConf = manager.getContext().getTableConfiguration(tableId);
MergeStats mergeStats = mergeStatsCache.computeIfAbsent(tableId, k -> {
var mStats = currentMerges.get(k);
return mStats != null ? mStats : new MergeStats(new MergeInfo());
});
TabletGoalState goal = manager.getGoalState(tls, mergeStats.getMergeInfo());
TServerInstance location = tls.getLocation();
TabletState state = tls.getState(currentTServers.keySet());
TabletLogger.missassigned(tls.extent, goal.toString(), state.toString(), tls.future, tls.current, tls.walogs.size());
stats.update(tableId, state);
mergeStats.update(tls.extent, state, tls.chopped, !tls.walogs.isEmpty());
sendChopRequest(mergeStats.getMergeInfo(), state, tls);
sendSplitRequest(mergeStats.getMergeInfo(), state, tls);
// Always follow through with assignments
if (state == TabletState.ASSIGNED) {
goal = TabletGoalState.HOSTED;
}
// if we are shutting down all the tabletservers, we have to do it in order
if ((goal == TabletGoalState.SUSPENDED && state == TabletState.HOSTED) && manager.serversToShutdown.equals(currentTServers.keySet())) {
if (dependentWatcher != null && dependentWatcher.assignedOrHosted() > 0) {
goal = TabletGoalState.HOSTED;
}
}
if (goal == TabletGoalState.HOSTED) {
if ((state != TabletState.HOSTED && !tls.walogs.isEmpty()) && manager.recoveryManager.recoverLogs(tls.extent, tls.walogs))
continue;
switch(state) {
case HOSTED:
if (location.equals(manager.migrations.get(tls.extent)))
manager.migrations.remove(tls.extent);
break;
case ASSIGNED_TO_DEAD_SERVER:
hostDeadTablet(tLists, tls, location, wals);
break;
case SUSPENDED:
hostSuspendedTablet(tLists, tls, location, tableConf);
break;
case UNASSIGNED:
hostUnassignedTablet(tLists, tls.extent, location);
break;
case ASSIGNED:
// Send another reminder
tLists.assigned.add(new Assignment(tls.extent, tls.future));
break;
}
} else {
switch(state) {
case SUSPENDED:
// Request a move to UNASSIGNED, so as to allow balancing to continue.
tLists.suspendedToGoneServers.add(tls);
cancelOfflineTableMigrations(tls.extent);
break;
case UNASSIGNED:
cancelOfflineTableMigrations(tls.extent);
break;
case ASSIGNED_TO_DEAD_SERVER:
unassignDeadTablet(tLists, tls, wals);
break;
case HOSTED:
TServerConnection client = manager.tserverSet.getConnection(location);
if (client != null) {
client.unloadTablet(manager.managerLock, tls.extent, goal.howUnload(), manager.getSteadyTime());
unloaded++;
totalUnloaded++;
} else {
Manager.log.warn("Could not connect to server {}", location);
}
break;
case ASSIGNED:
break;
}
}
counts[state.ordinal()]++;
}
flushChanges(tLists, wals);
// provide stats after flushing changes to avoid race conditions w/ delete table
stats.end(managerState);
// Report changes
for (TabletState state : TabletState.values()) {
int i = state.ordinal();
if (counts[i] > 0 && counts[i] != oldCounts[i]) {
manager.nextEvent.event("[%s]: %d tablets are %s", store.name(), counts[i], state.name());
}
}
Manager.log.debug(String.format("[%s]: scan time %.2f seconds", store.name(), stats.getScanTime() / 1000.));
oldCounts = counts;
if (totalUnloaded > 0) {
manager.nextEvent.event("[%s]: %d tablets unloaded", store.name(), totalUnloaded);
}
updateMergeState(mergeStatsCache);
synchronized (this) {
lastScanServers = ImmutableSortedSet.copyOf(currentTServers.keySet());
}
if (manager.tserverSet.getCurrentServers().equals(currentTServers.keySet())) {
Manager.log.debug(String.format("[%s] sleeping for %.2f seconds", store.name(), Manager.TIME_TO_WAIT_BETWEEN_SCANS / 1000.));
eventListener.waitForEvents(Manager.TIME_TO_WAIT_BETWEEN_SCANS);
} else {
Manager.log.info("Detected change in current tserver set, re-running state machine.");
}
} catch (Exception ex) {
Manager.log.error("Error processing table state for store " + store.name(), ex);
if (ex.getCause() != null && ex.getCause() instanceof BadLocationStateException) {
repairMetadata(((BadLocationStateException) ex.getCause()).getEncodedEndRow());
} else {
sleepUninterruptibly(Manager.WAIT_BETWEEN_ERRORS, TimeUnit.MILLISECONDS);
}
} finally {
if (iter != null) {
try {
iter.close();
} catch (IOException ex) {
Manager.log.warn("Error closing TabletLocationState iterator: " + ex, ex);
}
}
}
}
}
use of org.apache.accumulo.core.metadata.TabletLocationState in project accumulo by apache.
the class MergeStats method verifyMergeConsistency.
private boolean verifyMergeConsistency(AccumuloClient accumuloClient, CurrentState manager) throws TableNotFoundException, IOException {
MergeStats verify = new MergeStats(info);
KeyExtent extent = info.getExtent();
Scanner scanner = accumuloClient.createScanner(extent.isMeta() ? RootTable.NAME : MetadataTable.NAME, Authorizations.EMPTY);
MetaDataTableScanner.configureScanner(scanner, manager);
Text start = extent.prevEndRow();
if (start == null) {
start = new Text();
}
TableId tableId = extent.tableId();
Text first = TabletsSection.encodeRow(tableId, start);
Range range = new Range(first, false, null, true);
scanner.setRange(range.clip(TabletsSection.getRange()));
KeyExtent prevExtent = null;
log.debug("Scanning range {}", range);
for (Entry<Key, Value> entry : scanner) {
TabletLocationState tls;
try {
tls = MetaDataTableScanner.createTabletLocationState(entry.getKey(), entry.getValue());
} catch (BadLocationStateException e) {
log.error("{}", e.getMessage(), e);
return false;
}
log.debug("consistency check: {} walogs {}", tls, tls.walogs.size());
if (!tls.extent.tableId().equals(tableId)) {
break;
}
if (!tls.walogs.isEmpty() && verify.getMergeInfo().needsToBeChopped(tls.extent)) {
log.debug("failing consistency: needs to be chopped {}", tls.extent);
return false;
}
if (prevExtent == null) {
// the start of the merge range
if (tls.extent.prevEndRow() != null && tls.extent.prevEndRow().compareTo(start) > 0) {
log.debug("failing consistency: prev row is too high {}", start);
return false;
}
if (tls.getState(manager.onlineTabletServers()) != TabletState.UNASSIGNED && tls.getState(manager.onlineTabletServers()) != TabletState.SUSPENDED) {
log.debug("failing consistency: assigned or hosted {}", tls);
return false;
}
} else if (!tls.extent.isPreviousExtent(prevExtent)) {
log.debug("hole in {}", MetadataTable.NAME);
return false;
}
prevExtent = tls.extent;
verify.update(tls.extent, tls.getState(manager.onlineTabletServers()), tls.chopped, !tls.walogs.isEmpty());
// stop when we've seen the tablet just beyond our range
if (tls.extent.prevEndRow() != null && extent.endRow() != null && tls.extent.prevEndRow().compareTo(extent.endRow()) > 0) {
break;
}
}
log.debug("chopped {} v.chopped {} unassigned {} v.unassigned {} verify.total {}", chopped, verify.chopped, unassigned, verify.unassigned, verify.total);
return chopped == verify.chopped && unassigned == verify.unassigned && unassigned == verify.total;
}
use of org.apache.accumulo.core.metadata.TabletLocationState in project accumulo by apache.
the class RootTabletStateStoreTest method testRootTabletStateStore.
@Test
public void testRootTabletStateStore() throws DistributedStoreException {
ZooTabletStateStore tstore = new ZooTabletStateStore(new TestAmple());
KeyExtent root = RootTable.EXTENT;
String sessionId = "this is my unique session data";
TServerInstance server = new TServerInstance(HostAndPort.fromParts("127.0.0.1", 10000), sessionId);
List<Assignment> assignments = Collections.singletonList(new Assignment(root, server));
tstore.setFutureLocations(assignments);
int count = 0;
for (TabletLocationState location : tstore) {
assertEquals(location.extent, root);
assertEquals(location.future, server);
assertNull(location.current);
count++;
}
assertEquals(count, 1);
tstore.setLocations(assignments);
count = 0;
for (TabletLocationState location : tstore) {
assertEquals(location.extent, root);
assertNull(location.future);
assertEquals(location.current, server);
count++;
}
assertEquals(count, 1);
TabletLocationState assigned = null;
try {
assigned = new TabletLocationState(root, server, null, null, null, null, false);
} catch (BadLocationStateException e) {
fail("Unexpected error " + e);
}
tstore.unassign(Collections.singletonList(assigned), null);
count = 0;
for (TabletLocationState location : tstore) {
assertEquals(location.extent, root);
assertNull(location.future);
assertNull(location.current);
count++;
}
assertEquals(count, 1);
KeyExtent notRoot = new KeyExtent(TableId.of("0"), null, null);
final var assignmentList = List.of(new Assignment(notRoot, server));
assertThrows(IllegalArgumentException.class, () -> tstore.setLocations(assignmentList));
assertThrows(IllegalArgumentException.class, () -> tstore.setFutureLocations(assignmentList));
try {
TabletLocationState broken = new TabletLocationState(notRoot, server, null, null, null, null, false);
final var assignmentList1 = List.of(broken);
assertThrows(IllegalArgumentException.class, () -> tstore.unassign(assignmentList1, null));
} catch (BadLocationStateException e) {
fail("Unexpected error " + e);
}
}
Aggregations