Search in sources :

Example 1 with GlobalStateLogEntry

use of org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry in project nessie by projectnessie.

the class NonTransactionalDatabaseAdapter method transplant.

@SuppressWarnings("RedundantThrows")
@Override
public Hash transplant(BranchName targetBranch, Optional<Hash> expectedHead, List<Hash> sequenceToTransplant, Function<ByteString, ByteString> updateCommitMetadata) throws ReferenceNotFoundException, ReferenceConflictException {
    try {
        return casOpLoop("transplant", targetBranch, CasOpVariant.COMMIT, (ctx, pointer, branchCommits, newKeyLists) -> {
            Hash targetHead = branchHead(pointer, targetBranch);
            long timeInMicros = commitTimeInMicros();
            targetHead = transplantAttempt(ctx, timeInMicros, targetBranch, expectedHead, targetHead, sequenceToTransplant, branchCommits, newKeyLists, updateCommitMetadata);
            GlobalStateLogEntry newGlobalHead = writeGlobalCommit(ctx, timeInMicros, pointer, Collections.emptyList());
            RefLogEntry newRefLog = writeRefLogEntry(ctx, pointer, targetBranch.getName(), RefLogEntry.RefType.Branch, targetHead, RefLogEntry.Operation.TRANSPLANT, timeInMicros, sequenceToTransplant);
            // Return hash of last commit (targetHead) added to 'targetBranch' (via the casOpLoop)
            return updateGlobalStatePointer(targetBranch, pointer, targetHead, newGlobalHead, newRefLog);
        }, () -> transplantConflictMessage("Retry-failure", targetBranch, expectedHead, sequenceToTransplant));
    } catch (ReferenceNotFoundException | ReferenceConflictException | RuntimeException e) {
        throw e;
    } catch (Exception e) {
        throw new RuntimeException(e);
    }
}
Also used : ReferenceNotFoundException(org.projectnessie.versioned.ReferenceNotFoundException) GlobalStateLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry) ReferenceConflictException(org.projectnessie.versioned.ReferenceConflictException) DatabaseAdapterUtil.verifyExpectedHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.verifyExpectedHash) Hash(org.projectnessie.versioned.Hash) DatabaseAdapterUtil.randomHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.randomHash) RefLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.RefLogEntry) ReferenceAlreadyExistsException(org.projectnessie.versioned.ReferenceAlreadyExistsException) ReferenceConflictException(org.projectnessie.versioned.ReferenceConflictException) RefLogNotFoundException(org.projectnessie.versioned.RefLogNotFoundException) VersionStoreException(org.projectnessie.versioned.VersionStoreException) ReferenceNotFoundException(org.projectnessie.versioned.ReferenceNotFoundException)

Example 2 with GlobalStateLogEntry

use of org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry in project nessie by projectnessie.

the class NonTransactionalDatabaseAdapter method compactGlobalLog.

protected Map<String, String> compactGlobalLog(GlobalLogCompactionParams globalLogCompactionParams) {
    if (!globalLogCompactionParams.isEnabled()) {
        return ImmutableMap.of("compacted", "false", "reason", "not enabled");
    }
    // Not using casOpLoop() here, as it is simpler than adopting casOpLoop().
    try (TryLoopState tryState = newTryLoopState("compact-global-log", ts -> repoDescUpdateConflictMessage(String.format("%s after %d retries, %d ms", "Retry-Failure", ts.getRetries(), ts.getDuration(TimeUnit.MILLISECONDS))), this::tryLoopStateCompletion, config)) {
        CompactionStats stats = new CompactionStats();
        while (true) {
            NonTransactionalOperationContext ctx = NON_TRANSACTIONAL_OPERATION_CONTEXT;
            GlobalStatePointer pointer = fetchGlobalPointer(ctx);
            // Collect the old global-log-ids, to delete those after compaction
            List<Hash> oldLogIds = new ArrayList<>();
            // Map with all global contents.
            Map<String, ByteString> globalContents = new HashMap<>();
            // Content-IDs, most recently updated contents first.
            List<String> contentIdsByRecency = new ArrayList<>();
            // Read the global log - from the most recent global-log entry to the oldest.
            try (Stream<GlobalStateLogEntry> globalLog = globalLogFetcher(ctx, pointer)) {
                globalLog.forEach(e -> {
                    if (stats.read < globalLogCompactionParams.getNoCompactionWhenCompactedWithin() && stats.puts > stats.read) {
                        // do not compact.
                        throw COMPACTION_NOT_NECESSARY_WITHIN;
                    }
                    stats.read++;
                    oldLogIds.add(Hash.of(e.getId()));
                    for (ContentIdWithBytes put : e.getPutsList()) {
                        stats.puts++;
                        String cid = put.getContentId().getId();
                        if (globalContents.putIfAbsent(cid, put.getValue()) == null) {
                            stats.uniquePuts++;
                            contentIdsByRecency.add(cid);
                        }
                    }
                });
                if (stats.read < globalLogCompactionParams.getNoCompactionUpToLength()) {
                    // single-bulk read, so do not compact at all.
                    throw COMPACTION_NOT_NECESSARY_LENGTH;
                }
            } catch (RuntimeException e) {
                if (e == COMPACTION_NOT_NECESSARY_WITHIN) {
                    tryState.success(null);
                    return ImmutableMap.of("compacted", "false", "reason", String.format("compacted entry within %d most recent log entries", globalLogCompactionParams.getNoCompactionWhenCompactedWithin()));
                }
                if (e == COMPACTION_NOT_NECESSARY_LENGTH) {
                    tryState.success(null);
                    return ImmutableMap.of("compacted", "false", "reason", String.format("less than %d entries", globalLogCompactionParams.getNoCompactionUpToLength()));
                }
                throw e;
            }
            // Collect the IDs of the written global-log-entries, to delete those when the CAS
            // operation failed
            List<ByteString> newLogIds = new ArrayList<>();
            // Reverse the order of content-IDs, most recently updated contents LAST.
            // Do this to have the active contents closer to the HEAD of the global log.
            Collections.reverse(contentIdsByRecency);
            // Maintain the list of global-log-entry parent IDs, but in reverse order as in
            // GlobalLogEntry for easier management here.
            List<ByteString> globalParentsReverse = new ArrayList<>(config.getParentsPerGlobalCommit());
            globalParentsReverse.add(NO_ANCESTOR.asBytes());
            GlobalStateLogEntry.Builder currentEntry = newGlobalLogEntryBuilder(commitTimeInMicros()).addParents(globalParentsReverse.get(0));
            for (String cid : contentIdsByRecency) {
                if (currentEntry.buildPartial().getSerializedSize() >= config.getGlobalLogEntrySize()) {
                    compactGlobalLogWriteEntry(ctx, stats, globalParentsReverse, currentEntry, newLogIds);
                    // Prepare new entry
                    currentEntry = newGlobalLogEntryBuilder(commitTimeInMicros());
                    for (int i = globalParentsReverse.size() - 1; i >= 0; i--) {
                        currentEntry.addParents(globalParentsReverse.get(i));
                    }
                }
                ByteString value = globalContents.get(cid);
                currentEntry.addPuts(ContentIdWithBytes.newBuilder().setContentId(AdapterTypes.ContentId.newBuilder().setId(cid)).setTypeUnused(0).setValue(value).build());
            }
            compactGlobalLogWriteEntry(ctx, stats, globalParentsReverse, currentEntry, newLogIds);
            GlobalStatePointer newPointer = GlobalStatePointer.newBuilder().addAllNamedReferences(pointer.getNamedReferencesList()).addAllRefLogParentsInclHead(pointer.getRefLogParentsInclHeadList()).setRefLogId(pointer.getRefLogId()).setGlobalId(currentEntry.getId()).addGlobalParentsInclHead(currentEntry.getId()).addAllGlobalParentsInclHead(currentEntry.getParentsList()).build();
            stats.addToTotal();
            // CAS global pointer
            if (globalPointerCas(ctx, pointer, newPointer)) {
                tryState.success(null);
                cleanUpGlobalLog(ctx, oldLogIds);
                return stats.asMap(tryState);
            }
            // Note: if it turns out that there are too many CAS retries happening, the overall
            // mechanism can be updated as follows. Since the approach below is much more complex
            // and harder to test, if's not part of the initial implementation.
            // 
            // 1. Read the whole global-log as currently, but outside the actual CAS-loop.
            // Save the current HEAD of the global-log
            // 2. CAS-loop:
            // 2.1. Construct and write the new global-log
            // 2.2. Try the CAS, if it succeeds, fine
            // 2.3. If the CAS failed:
            // 2.3.1. Clean up the optimistically written new global-log
            // 2.3.2. Read the global-log from its new HEAD up to the current HEAD from step 1.
            // Only add the most-recent values for the content-IDs in the incrementally
            // read global-log
            // 2.3.3. Remember the "new HEAD" as the "current HEAD"
            // 2.3.4. Continue to step 2.1.
            cleanUpGlobalLog(ctx, newLogIds.stream().map(Hash::of).collect(Collectors.toList()));
            stats.onRetry();
            tryState.retry();
        }
    } catch (ReferenceConflictException e) {
        throw new RuntimeException(e);
    }
}
Also used : GlobalStatePointer(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStatePointer) GlobalStateLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry) HashMap(java.util.HashMap) ByteString(com.google.protobuf.ByteString) ReferenceConflictException(org.projectnessie.versioned.ReferenceConflictException) ArrayList(java.util.ArrayList) ByteString(com.google.protobuf.ByteString) DatabaseAdapterUtil.verifyExpectedHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.verifyExpectedHash) Hash(org.projectnessie.versioned.Hash) DatabaseAdapterUtil.randomHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.randomHash) Builder(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry.Builder) TryLoopState.newTryLoopState(org.projectnessie.versioned.persist.adapter.spi.TryLoopState.newTryLoopState) TryLoopState(org.projectnessie.versioned.persist.adapter.spi.TryLoopState) ContentIdWithBytes(org.projectnessie.versioned.persist.serialize.AdapterTypes.ContentIdWithBytes)

Example 3 with GlobalStateLogEntry

use of org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry in project nessie by projectnessie.

the class NonTransactionalDatabaseAdapter method merge.

@Override
public Hash merge(Hash from, BranchName toBranch, Optional<Hash> expectedHead, Function<ByteString, ByteString> updateCommitMetadata) throws ReferenceNotFoundException, ReferenceConflictException {
    // creates a new commit-tree that is decoupled from other commit-trees.
    try {
        return casOpLoop("merge", toBranch, CasOpVariant.COMMIT, (ctx, pointer, branchCommits, newKeyLists) -> {
            Hash toHead = branchHead(pointer, toBranch);
            long timeInMicros = commitTimeInMicros();
            toHead = mergeAttempt(ctx, timeInMicros, from, toBranch, expectedHead, toHead, branchCommits, newKeyLists, updateCommitMetadata);
            GlobalStateLogEntry newGlobalHead = writeGlobalCommit(ctx, timeInMicros, pointer, Collections.emptyList());
            RefLogEntry newRefLog = writeRefLogEntry(ctx, pointer, toBranch.getName(), RefLogEntry.RefType.Branch, toHead, RefLogEntry.Operation.MERGE, timeInMicros, Collections.singletonList(from));
            // Return hash of last commit (toHead) added to 'targetBranch' (via the casOpLoop)
            return updateGlobalStatePointer(toBranch, pointer, toHead, newGlobalHead, newRefLog);
        }, () -> mergeConflictMessage("Retry-failure", from, toBranch, expectedHead));
    } catch (ReferenceNotFoundException | ReferenceConflictException | RuntimeException e) {
        throw e;
    } catch (Exception e) {
        throw new RuntimeException(e);
    }
}
Also used : ReferenceNotFoundException(org.projectnessie.versioned.ReferenceNotFoundException) GlobalStateLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry) ReferenceConflictException(org.projectnessie.versioned.ReferenceConflictException) DatabaseAdapterUtil.verifyExpectedHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.verifyExpectedHash) Hash(org.projectnessie.versioned.Hash) DatabaseAdapterUtil.randomHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.randomHash) RefLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.RefLogEntry) ReferenceAlreadyExistsException(org.projectnessie.versioned.ReferenceAlreadyExistsException) ReferenceConflictException(org.projectnessie.versioned.ReferenceConflictException) RefLogNotFoundException(org.projectnessie.versioned.RefLogNotFoundException) VersionStoreException(org.projectnessie.versioned.VersionStoreException) ReferenceNotFoundException(org.projectnessie.versioned.ReferenceNotFoundException)

Example 4 with GlobalStateLogEntry

use of org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry in project nessie by projectnessie.

the class NonTransactionalDatabaseAdapter method writeGlobalCommit.

protected GlobalStateLogEntry writeGlobalCommit(NonTransactionalOperationContext ctx, long timeInMicros, GlobalStatePointer pointer, List<ContentIdAndBytes> globals) throws ReferenceConflictException {
    Hash parentHash = Hash.of(pointer.getGlobalId());
    // Before Nessie 0.21.0: the global-state-pointer contains the "heads" for the ref-log and
    // global-log, so it has to read the head ref-log & global-log to get the IDs of all the
    // previous parents to fill the parents in the new global-log-entry & ref-log-entry.
    // 
    // Since Nessie 0.21.0: the global-state-pointer contains the "heads" for the ref-log and
    // global-log PLUS the parents of those, so Nessie no longer need to read the head entries
    // from the ref-log + global-log.
    // 
    // The check of the first entry is there to ensure backwards compatibility and also
    // rolling-upgrades work.
    Stream<ByteString> newParents;
    if (pointer.getGlobalParentsInclHeadCount() == 0 || !pointer.getGlobalId().equals(pointer.getGlobalParentsInclHead(0))) {
        // Before Nessie 0.21.0
        newParents = Stream.of(parentHash.asBytes());
        GlobalStateLogEntry currentEntry = fetchFromGlobalLog(ctx, parentHash);
        if (currentEntry != null) {
            newParents = Stream.concat(newParents, currentEntry.getParentsList().stream().limit(config.getParentsPerGlobalCommit() - 1));
        }
    } else {
        // Since Nessie 0.21.0
        newParents = pointer.getGlobalParentsInclHeadList().stream().limit(config.getParentsPerGlobalCommit());
    }
    GlobalStateLogEntry.Builder entry = newGlobalLogEntryBuilder(timeInMicros);
    newParents.forEach(entry::addParents);
    globals.forEach(g -> entry.addPuts(ProtoSerialization.toProto(g)));
    GlobalStateLogEntry globalLogEntry = entry.build();
    writeGlobalCommit(ctx, globalLogEntry);
    return globalLogEntry;
}
Also used : Builder(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry.Builder) GlobalStateLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry) ByteString(com.google.protobuf.ByteString) DatabaseAdapterUtil.verifyExpectedHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.verifyExpectedHash) Hash(org.projectnessie.versioned.Hash) DatabaseAdapterUtil.randomHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.randomHash)

Example 5 with GlobalStateLogEntry

use of org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry in project nessie by projectnessie.

the class NonTransactionalDatabaseAdapter method create.

@Override
public Hash create(NamedRef ref, Hash target) throws ReferenceAlreadyExistsException, ReferenceNotFoundException {
    try {
        return casOpLoop("createRef", ref, CasOpVariant.REF_UPDATE, (ctx, pointer, branchCommits, newKeyLists) -> {
            if (refFromGlobalState(pointer, ref.getName()) != null) {
                throw referenceAlreadyExists(ref);
            }
            Hash hash = target;
            if (hash == null) {
                // Special case: Don't validate, if the 'target' parameter is null.
                // This is mostly used for tests that re-create the default-branch.
                hash = NO_ANCESTOR;
            }
            validateHashExists(ctx, hash);
            // Need a new empty global-log entry to be able to CAS
            GlobalStateLogEntry newGlobalHead = noopGlobalLogEntry(ctx, pointer);
            RefLogEntry.RefType refType = ref instanceof TagName ? RefLogEntry.RefType.Tag : RefLogEntry.RefType.Branch;
            RefLogEntry newRefLog = writeRefLogEntry(ctx, pointer, ref.getName(), refType, hash, RefLogEntry.Operation.CREATE_REFERENCE, commitTimeInMicros(), Collections.emptyList());
            return updateGlobalStatePointer(ref, pointer, hash, newGlobalHead, newRefLog);
        }, () -> createConflictMessage("Retry-Failure", ref, target));
    } catch (ReferenceAlreadyExistsException | ReferenceNotFoundException | RuntimeException e) {
        throw e;
    } catch (Exception e) {
        throw new RuntimeException(e);
    }
}
Also used : ReferenceNotFoundException(org.projectnessie.versioned.ReferenceNotFoundException) GlobalStateLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry) RefType(org.projectnessie.versioned.persist.serialize.AdapterTypes.RefLogEntry.RefType) TagName(org.projectnessie.versioned.TagName) ReferenceAlreadyExistsException(org.projectnessie.versioned.ReferenceAlreadyExistsException) DatabaseAdapterUtil.verifyExpectedHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.verifyExpectedHash) Hash(org.projectnessie.versioned.Hash) DatabaseAdapterUtil.randomHash(org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.randomHash) RefLogEntry(org.projectnessie.versioned.persist.serialize.AdapterTypes.RefLogEntry) ReferenceAlreadyExistsException(org.projectnessie.versioned.ReferenceAlreadyExistsException) ReferenceConflictException(org.projectnessie.versioned.ReferenceConflictException) RefLogNotFoundException(org.projectnessie.versioned.RefLogNotFoundException) VersionStoreException(org.projectnessie.versioned.VersionStoreException) ReferenceNotFoundException(org.projectnessie.versioned.ReferenceNotFoundException)

Aggregations

GlobalStateLogEntry (org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStateLogEntry)12 Hash (org.projectnessie.versioned.Hash)11 DatabaseAdapterUtil.randomHash (org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.randomHash)11 ReferenceConflictException (org.projectnessie.versioned.ReferenceConflictException)10 DatabaseAdapterUtil.verifyExpectedHash (org.projectnessie.versioned.persist.adapter.spi.DatabaseAdapterUtil.verifyExpectedHash)10 RefLogEntry (org.projectnessie.versioned.persist.serialize.AdapterTypes.RefLogEntry)10 RefLogNotFoundException (org.projectnessie.versioned.RefLogNotFoundException)8 ReferenceAlreadyExistsException (org.projectnessie.versioned.ReferenceAlreadyExistsException)8 ReferenceNotFoundException (org.projectnessie.versioned.ReferenceNotFoundException)8 VersionStoreException (org.projectnessie.versioned.VersionStoreException)8 ByteString (com.google.protobuf.ByteString)5 TagName (org.projectnessie.versioned.TagName)5 RefType (org.projectnessie.versioned.persist.serialize.AdapterTypes.RefLogEntry.RefType)5 CommitLogEntry (org.projectnessie.versioned.persist.adapter.CommitLogEntry)4 GlobalStatePointer (org.projectnessie.versioned.persist.serialize.AdapterTypes.GlobalStatePointer)4 ArrayList (java.util.ArrayList)3 Collections (java.util.Collections)3 HashMap (java.util.HashMap)3 HashSet (java.util.HashSet)3 Set (java.util.Set)3