agena

This release has operational improvements, to the lua filters and the self-telemetry that cernan is able to produce. The big change here is to the file subsystem which is not more responsive to fast files and consumes less CPU time when changes to files are slow.

• 62a49d4 :: Allow lua filters to require other lua modules
• 3b1fbd2 :: Increase the responsiveness of the file source
• f26a094 :: Introduce the 'internal' source
• f2769b8 :: Make sampling inclusive of all statsd types
• a963db3 :: Increase existing telemetry, adjust file delays

agena

This release has a number of small fixes as well as improvements to the filter subsystem. The changes are:

• a0c5814 :: Config parse errors are now more clear
• 77fd9d2 :: Graphite/Statsd may now be bound to DNS hostname
• 9dbf4c6 :: Wavefront sink percentiles are now configurable
• 9148b5e :: Filter payloads can now be cleared of all logs / metrics
• dd0093c :: Internal ticks are 1/sec, filters receive ticks properly

Each individual commit contains further details. The wiki will be updated shortly as appropriate. Heavy users of filters are recommended to upgrade.

agena

This release correct an issue with file source whereby empty globs will consume large amounts of CPU. This does not impact systems where the file source has files to consume but is a recommended upgraded. It will not impact systems that do not use the file source.

agena

This release updates the hopper dependency to 0.2.1 and does some housekeeping to make it possible for us to release cernan on crates.io. The bump to 0.2.1 resolves a build issue caused by the dependency of hopper on bincode 1.0.0-alpha1.

agena

This release incorporates repairs to the firehose sink which, subsequent to 0.4.12, was accidentally nerfed. That is, the sink was no longer able to store delivered lines into its queue. This release also updates to serde 0.9, making rust 1.15 the minimum rust version that can build cernan. We have also updated to protobuf 1.2, reducing some of the cloning that goes on around the native protocol source/sink.

agena

This release includes documentation changes, some performance work around the native protocol and a fix to the sampling behaviour in the statsd protocol. It is the first release to include community commits: very exciting! The changes are as follows:

• d8944d9 :: CI coverage is enabled
• a88f95c :: outdated documentation removed
• 442d977 :: typo in documentation corrected
• af12873 :: rust lints introduced, WIP on missing-docs
• 4941129 :: native protocol impl. now make less temp. allocs
• 4a6ef95 :: project deps updated
• 032e0ba :: b_spec statsd sampling syntax supported

Signed-off-by: Brian L. Troutwine blt@postmates.com

agena

This release is a long time coming and closes the venerable 0.4.x
series of development. Cernan has been put on a good foot for
being open-sourced, has received significant optimizations and
been given three new sinks and one new source. The changes are
as follows:

• ccced68 :: Telemetry now avoids allocing a CKMS if not needed
• 3d88add :: mpsc is split out into the open-source hopper
• 3937bc4 :: native protocol is introduced, Arc sprinkled throughout
• 22668f4 :: Experimental InfluxDB sink is introduced
• 4746766 :: Numerous small allocations in text sources are removed
• f48403a :: Experimental Prometheus sink is introduced
• 4b609f3 :: Cernan's internal data-model becomes that of native protocol
• b1d482f :: Aggregation optimizations, wrt cache access

There were a few things I skipped in the above either because they
were later supplanted or because it was hard to fit the summary in
a short line. But that's the gist!

Signed-off-by: Brian L. Troutwine blt@postmates.com

bellwood

This release includes a grab bag of performance optimizations, including far fewer allocations in the programmable filter. As well, we fix my new favorite bug. See the message on commit 3692932 for details.

bellwood

This release stabilizes the programmable filters, making it possible to manipulate telemetry and logs in-flight with user-written lua code. This is a big accomplishment, one of the long-term goals of cernan is now achieved.

This release also fixes a bug in the interpretation of gauges by making a hard distinction between delta-gauges and gauges. Previously their individual behaviours were somewhat confused.

Lastly, this release introduces some performance improvements driven by callgrind profiling of cernan. There's more that can be done--especially around metrics which are mostly the same internally--but this was all low-hanging fruit.

bellwood

This release fixes a long-standing and never before recognized issue with the way gauges' times were reset on bucket reset. Previously, the last gauge's time was retained across a reset until a new point came in. This works for relatively well-trafficed series but does not hold up for sparse series of DeltaGauges.

We now reset times on bucket resets.