rename new sstables with correct generation at load time patch by jbellis; reviewed by slebresne for CASSANDRA-3967
Project: http://git-wip-us.apache.org/repos/asf/cassandra/repo Commit: http://git-wip-us.apache.org/repos/asf/cassandra/commit/837ee0ec Tree: http://git-wip-us.apache.org/repos/asf/cassandra/tree/837ee0ec Diff: http://git-wip-us.apache.org/repos/asf/cassandra/diff/837ee0ec Branch: refs/heads/cassandra-1.1.0 Commit: 837ee0ec5659afae2d3f138ee55d3bc7a60bdebf Parents: 6423bfe Author: Jonathan Ellis <jbel...@apache.org> Authored: Thu Mar 8 14:00:47 2012 -0600 Committer: Jonathan Ellis <jbel...@apache.org> Committed: Fri Mar 16 11:20:33 2012 -0500 ---------------------------------------------------------------------- .../org/apache/cassandra/db/ColumnFamilyStore.java | 93 ++++---------- .../apache/cassandra/io/sstable/SSTableWriter.java | 7 +- 2 files changed, 33 insertions(+), 67 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/cassandra/blob/837ee0ec/src/java/org/apache/cassandra/db/ColumnFamilyStore.java ---------------------------------------------------------------------- diff --git a/src/java/org/apache/cassandra/db/ColumnFamilyStore.java b/src/java/org/apache/cassandra/db/ColumnFamilyStore.java index ef9bc90..7bc3c95 100644 --- a/src/java/org/apache/cassandra/db/ColumnFamilyStore.java +++ b/src/java/org/apache/cassandra/db/ColumnFamilyStore.java @@ -449,75 +449,59 @@ public class ColumnFamilyStore implements ColumnFamilyStoreMBean { logger.info("Loading new SSTables for " + table.name + "/" + columnFamily + "..."); - // current view over ColumnFamilyStore - DataTracker.View view = data.getView(); - // descriptors of currently registered SSTables Set<Descriptor> currentDescriptors = new HashSet<Descriptor>(); - // going to hold new SSTable view of the CFS containing old and new SSTables - Set<SSTableReader> sstables = new HashSet<SSTableReader>(); - // get the max generation number, to prevent generation conflicts - int generation = 0; - - for (SSTableReader reader : view.sstables) - { - sstables.add(reader); // first of all, add old SSTables - currentDescriptors.add(reader.descriptor); - - if (reader.descriptor.generation > generation) - generation = reader.descriptor.generation; - } - - SSTableReader reader; - // set to true if we have at least one new SSTable to load - boolean atLeastOneNew = false; + for (SSTableReader sstable : data.getView().sstables) + currentDescriptors.add(sstable.descriptor); + Set<SSTableReader> newSSTables = new HashSet<SSTableReader>(); Directories.SSTableLister lister = directories.sstableLister().skipCompacted(true).skipTemporary(true); - for (Map.Entry<Descriptor, Set<Component>> rawSSTable : lister.list().entrySet()) + for (Map.Entry<Descriptor, Set<Component>> entry : lister.list().entrySet()) { - Descriptor descriptor = rawSSTable.getKey(); + Descriptor descriptor = entry.getKey(); if (currentDescriptors.contains(descriptor)) continue; // old (initialized) SSTable found, skipping + if (descriptor.temporary) // in the process of being written + continue; if (!descriptor.isCompatible()) throw new RuntimeException(String.format("Can't open incompatible SSTable! Current version %s, found file: %s", Descriptor.CURRENT_VERSION, descriptor)); - logger.info("Initializing new SSTable {}", rawSSTable); + Descriptor newDescriptor = new Descriptor(descriptor.directory, + descriptor.ksname, + descriptor.cfname, + fileIndexGenerator.incrementAndGet(), + false); + logger.info("Renaming new SSTable {} to {}", descriptor, newDescriptor); + SSTableWriter.rename(descriptor, newDescriptor, entry.getValue()); + SSTableReader reader; try { - Set<DecoratedKey> savedKeys = CacheService.instance.keyCache.readSaved(descriptor.ksname, descriptor.cfname); - reader = SSTableReader.open(rawSSTable.getKey(), rawSSTable.getValue(), savedKeys, data, metadata, partitioner); + reader = SSTableReader.open(newDescriptor, entry.getValue(), Collections.<DecoratedKey>emptySet(), data, metadata, partitioner); } catch (IOException e) { - SSTableReader.logOpenException(rawSSTable.getKey(), e); + SSTableReader.logOpenException(entry.getKey(), e); continue; } - - sstables.add(reader); - - if (descriptor.generation > generation) - generation = descriptor.generation; - - if (!atLeastOneNew) // set flag only once - atLeastOneNew = true; + newSSTables.add(reader); } - if (!atLeastOneNew) + if (newSSTables.isEmpty()) { logger.info("No new SSTables where found for " + table.name + "/" + columnFamily); return; } - logger.info("Loading new SSTables and building secondary indexes for " + table.name + "/" + columnFamily + ": " + sstables); - SSTableReader.acquireReferences(sstables); - data.addSSTables(sstables); + logger.info("Loading new SSTables and building secondary indexes for " + table.name + "/" + columnFamily + ": " + newSSTables); + SSTableReader.acquireReferences(newSSTables); + data.addSSTables(newSSTables); try { - indexManager.maybeBuildSecondaryIndexes(sstables, indexManager.getIndexedColumns()); + indexManager.maybeBuildSecondaryIndexes(newSSTables, indexManager.getIndexedColumns()); } catch (IOException e) { @@ -525,20 +509,7 @@ public class ColumnFamilyStore implements ColumnFamilyStoreMBean } finally { - SSTableReader.releaseReferences(sstables); - } - - if (fileIndexGenerator.get() < generation) - { - // we don't bother with CAS here since if the generations used in the new files overlap with - // files that we create during load, we're already screwed - logger.info("Setting up new generation: " + generation); - fileIndexGenerator.set(generation); - } - else - { - logger.warn("Largest generation seen in loaded sstables was {}, which may overlap with native sstable files (generation {}).", - generation, fileIndexGenerator.get()); + SSTableReader.releaseReferences(newSSTables); } logger.info("Done loading load new SSTables for " + table.name + "/" + columnFamily); @@ -1621,22 +1592,12 @@ public class ColumnFamilyStore implements ColumnFamilyStoreMBean } /** - * For testing. no effort is made to clear historical memtables, nor for - * thread safety + * For testing. No effort is made to clear historical or even the current memtables, nor for + * thread safety. All we do is wipe the sstable containers clean, while leaving the actual + * data files present on disk. (This allows tests to easily call loadNewSSTables on them.) */ public void clearUnsafe() { - fileIndexGenerator.set(0); // Avoid unit test failures (see CASSANDRA-3735). - - // Clear backups - Directories.SSTableLister lister = directories.sstableLister().onlyBackups(true); - for (Map.Entry<Descriptor, Set<Component>> entry : lister.list().entrySet()) - { - Descriptor desc = entry.getKey(); - for (Component comp : entry.getValue()) - FileUtils.delete(desc.filenameFor(comp)); - } - for (ColumnFamilyStore cfs : concatWithIndexes()) cfs.data.init(); } http://git-wip-us.apache.org/repos/asf/cassandra/blob/837ee0ec/src/java/org/apache/cassandra/io/sstable/SSTableWriter.java ---------------------------------------------------------------------- diff --git a/src/java/org/apache/cassandra/io/sstable/SSTableWriter.java b/src/java/org/apache/cassandra/io/sstable/SSTableWriter.java index 25c158c..5e99241 100644 --- a/src/java/org/apache/cassandra/io/sstable/SSTableWriter.java +++ b/src/java/org/apache/cassandra/io/sstable/SSTableWriter.java @@ -363,6 +363,12 @@ public class SSTableWriter extends SSTable static Descriptor rename(Descriptor tmpdesc, Set<Component> components) { Descriptor newdesc = tmpdesc.asTemporary(false); + rename(tmpdesc, newdesc, components); + return newdesc; + } + + public static void rename(Descriptor tmpdesc, Descriptor newdesc, Set<Component> components) + { try { // do -Data last because -Data present should mean the sstable was completely renamed before crash @@ -374,7 +380,6 @@ public class SSTableWriter extends SSTable { throw new IOError(e); } - return newdesc; } public long getFilePointer()
