v1.0.0

This release represents a major rewrite of sequins. We've been running this new code internally for more than six months, and we think it's in pretty good shape!

The main difference is that sequins can now run as a cluster, partitioning and replicating data across multiple machines. There are lots of other changes, as well, including:

• A better on-disk format, including on-disk compression
• File-based configuration, with lots of configuration options
• A web interface with status information, and lots of monitoring hooks
• A comprehensive manual

To say the least, this release is highly backwards incompatible with sequins 0.4.x. Moreover, input data that was organized for 0.4.x needs to be organized slightly different in S3 or HDFS. You can read more about the new source root layout here.

Additionally, since v1.0.0rc1, a few issues have been fixed:

• #60: A rare deadlock in the zookeeper code was fixed, and automatic GCing of old zookeeper nodes was added
• #59: A bug was fixed where sequins would cancel in-flight proxied requests
• #58: Sequins can now start up with no data, to make the quickstart guide simpler

v1.0.0rc1

This is the first draft release of v1.0, a major rewrite of sequins. We've been running this new code internally for more than six months, and we think it's in pretty good shape!

The main difference is that sequins can now run as a cluster, partitioning and replicating data across multiple machines. There are lots of other changes, as well, including:

• A better on-disk format, including on-disk compression
• File-based configuration, with lots of configuration options
• A web interface with status information, and lots of monitoring hooks
• A comprehensive manual

To say the least, this release is highly backwards incompatible with sequins 0.4.x. Moreover, input data that was organized for 0.4.x needs to be organized slightly different in S3 or HDFS. You can read more about the new source root layout here.

v0.4.7

This minor release fixes a bug in the previous release - downloading data to temporary directories before renaming.

It also includes a new utility, sequins-dump, to dump out the data in sequencefiles. You use it like this:

$ sequins-dump test_data/0/
Alice     Practice
Bob       Hope
Charlie   Horse

It also has options to dump any combination of offsets, keys, and values:

$ sequins-dump --help
usage: sequins-dump [<flags>] <PATH>

Flags:
  --help         Show help (also see --help-long and --help-man).
  -o, --offsets  Display offsets.
  -k, --keys     Display keys.
  -v, --values   Display values.
  --version      Show application version.

v0.4.6

This point release contains one minor change: instead of downloading data directly into the destination local dir, sequins will first download to a temporary dir and then rename the directory into place.

v0.4.5

This fixes the previous broken release where the index would get unset immediately after starting the HTTP server, causing 500's to return when trying to make HTTP requests.

v0.4.4

This version brings two changes:

1. A bugfix to remove a race condition which could cause requests to fail when hotloading a new version of the dataset.
2. Add a X-Sequins-Version header to responses, which indicates the current version of the dataset.

v0.4.3

This minor release brings two small changes:

• A bug has been fixed where sequins would keep indexed files open long after it had switched to a new version.
• Sequins now writes out a .manifest file once it has fully indexed a dataset, and it can load from that on startup instead of indexing everything all over again.

v0.4.2

This is another bugfix release, for a second regression introduced in v0.4.0. Because of a copied lock, some rare race conditions were introduced to the read path.

v0.4.1

This is a hotfix release for a regression introduced in 0.4.1. In that version, the sync marker in sequencefiles could sometimes be read incorrectly, causing sequins to think that the file was corrupted.

v0.4.0

This release comes with some performance optimizations, as well as some code cleanup. The two major changes are:

• Reading SequenceFiles now requires significantly less heap bloat; this should reduce memory usage for large datasets considerably
• Indexing now happens in serial, instead of in parallel - this means it's slightly slower, but thrashes the disk a lot less.