Shackle is a horizontally scalable distributed hash set with automatic eviction, linearizable isolation and two phase commit built on LMDB and Dragonboat. It was created specifically to provide a durable, consistent, cost efficient solution to high volume event stream deduplication.
The API consists of 3 operations available via JSON API or GRPC. All operations take an array of strings (or bytes) and return an array of integers. Each response integer represents a status code for the corresponding item in the batch.
Lock [“aaa”, “bbb”, “ccc”, “ddd”, “eee”, “fff”]
Response: [2, 2, 2, 2, 3, 4]
Four were successfully locked (2) and will remain so for 30 seconds.
One item was busy (3), and one already exists (4) in the database.
Rollback [“aaa”]
Response: [1]
Here, one previously locked item is successfully rolled back to open (1).
This is common (and necessary) when a client fails to process an item.
Commit [“bbb”, “cccc”, “dddd”]
Response: [4, 4, 4]
These three locked items are committed to the database and show as exists (4).
They will remain committed until they expire.
These three simple batch operations can be combined to implement an observable message deduplication system where duplicate delivery can be drastically reduced and exactly once delivery can be measured accurately.
The terminology used by this library is slightly unconventional. Rather than following conventional cluster/node/partition terminology, it borrows its deployment/host/node terminology from its current raft integration (dragonboat).
Clone the source and run make dev to spin up a 3 node local cluster. Open _examples/local_3x2/config.*.yml to inspect and alter the expected directory configuration.
This repository is in pre-alpha. It has been made public for review. The master branch should work, but should not be deployed to staging environments until tagged v0.1.0
- No migrations.
- No discovery.
- No recovery.
- No auth.
- Static pepper.
- Fixed expiration.
- HTTP/RPC interface.
- Lock propagation.
- Lock timeout.
- Rollback.
- Commit.
- Buckets.
Journaling is enabled by default in xfs, ext3, ext4 and ntfs but it is a setting that might be disabled.
Use of a writeback filesystem (the opposite of ordered/journaled) can cause database corruption on system crash because writeback filesystems sometimes write pages to disk out of order while journaled filesystems do not.
You can use debugfs to ensure that your filesystem is configured with has_journal enabled.
> debugfs -R features /dev/sda1
debugfs 1.42.9 (28-Dec-2013)
Filesystem features: has_journal ext_attr resize_inode dir_index filetype needs_recovery extent 64bit
flex_bg sparse_super large_file huge_file uninit_bg dir_nlink extra_isize