Journal.IO is a lightweight, fast and easy-to-use journal storage implementation based on append-only rotating logs and checksummed variable-length records, supporting concurrent reads and writes, dynamic batching, tunable durability and data compaction.
Journal.IO has been forked from the HawtJournal project, in order to provide new features and fixes with a faster development and release cycle, as well as foster open collaboration.
APIs are very simple and intuitive.
First, create and configure the journal:
Journal journal = new Journal();
journal.setDirectory(JOURNAL_DIR);
Then, open the journal:
journal.open();
And write some records:
for (int i = 0; i < writes; i++) {
boolean sync = i % 2 == 0 ? true : false;
journal.write(new String("DATA" + i), WriteType.SYNC);
}
You can dynamically write either in async (WriteType.ASYNC) or sync mode (WriteType.SYNC): in async mode, writes are batched until either the max batch size is reached, the journal is manually synced or closed, or a sync write is executed.
The, forward-replay the log by going through the Journal "redo" iterable, obtaining record locations by reading them either in async (ReadType.ASYNC) or sync mode (ReadType.SYNC): async mode takes advantage of speculative reads and so it's faster than sync mode, which instead is slower but able to suddenly detect deleted records:
for (Location location : journal.redo()) {
byte[] record = journal.read(location, ReadType.SYNC);
// do something
}
You can also backward-replay the log by going through the Journal "undo" iterable:
for (Location location : journal.undo()) {
byte[] record = journal.read(location, ReadType.SYNC);
// do something
}
Eventually delete some record:
journal.delete(location);
Optionally do a manual sync:
journal.sync();
Compact logs:
journal.compact();
And close it:
journal.close();
That's all!
Journal.IO is thread safe: this means the same Journal instance can be concurrently accessed by multiple threads, for both reading and writing, inside the same JVM process. By the way, you cannot access the same set of journal files from different JVM processes: future Journal.IO versions will implement a simple file locking mechanism to prevent data corruption due to shared file access.
Journal.IO provides its own managed threads: exactly one thread for writing batches, and one thread for disposing unused resources, so it is really cheap in terms of thread usage. However, you can provide your own managed executors by setting the Journal#setWriter and Journal#setDisposer methods: by doing so, Journal.IO will use the provided executors and hence share threads with other parts of your application; please note this should be done only if you really know what you are doing.
Journal.IO provides three levels of data durability: batch, sync and physical sync.
Batch durability provides the lowest durability guarantees but the greatest performance: data is collected in memory batches and then written on disk at "sync points", either when a sync write is explicitly requested, a sync call is explicitly executed, or when the max batch size is reached.
Sync durability happens during sync points, providing higher durability guarantees with some performance cost: here, memory batches are written on disk.
Finally, physical sync durability provides the highest durability guarantees at the expense of a greater performance cost, flushing on disk hardware buffers at every sync point: disabled by default, it can be enabled by setting Journal#setPhysicalSync true.
You can download Journal.IO via Maven repositories, here are the coordinates:
<dependency>
<groupId>com.github.sbtourist</groupId>
<artifactId>journalio</artifactId>
<version>1.3</version>
</dependency>
- Journal.IO Migration tool: to migrate from Journal.IO 1.2 to 1.3 log format.
Join the mailing list at: http://groups.google.com/group/journalio.
Or feel free to contact me on Twitter.
Distributed under the Apache Software License.
Journal.IO is based on the HawtJournal project, for original copyright note see: HawtJournal.
Sergio Bossa (Journal.IO maintainer).
Hiram Chirino (Original author of the HawtJournal project).
Giuseppe Santoro (Contributor)
Chris Vest (Contributor)
Martin Krasser (Contributor)
Holger Hoffstätte (Contributor)
Julien Eluard (Contributor)
Brian McCallister (Contributor)