Skip to main content

Releasing 0.9

· 3 min read
The LittleHorse Council
The LittleHorse Council
The Council of LittleHorse Maintainers

The 0.9.2 release is now availble and ready for use. The 0.9.x releases focused mainly on:

  • Improving the user experience on the LittleHorse Dashboard
  • Improving the reliability of the LH Server in the face of rebalances and failures.

New Features​

While the majority of the improvements in the 0.9 release revolve around performance and stability, several of them are highly visible to the user (especially the new dashboard!).

Dashboard Rewrite​

With help from Nelson Jumbo, LittleHorse Knight Mijaíl Rondón rewrote and revamped our administrative dashboard. It now inclues new features such as:

  • User Task Detail page
  • Improved details on TaskRun progress
  • Improved details on WfRun progress
  • A plethora of small bug fixes.

Internal Task Queue Optimizations​

Deep in the internals of the LittleHorse Server, we implement a Task Queue mechanism to store ScheduledTasks before they're dispatched to the Task Worker clients. This release included many improvements to stability of the Task Queues.

Most importantly, our Grumpy Maintainer (Eduwer Camacaro) put a cap on the memory consumption of a single TaskDef. Prior to this release, it was possible for poorly-behaved clients to cause an OOM on the server by running millions of workflows which use a TaskDef but not executing the resulting TaskRuns. This would cause an un-bounded buildup of ScheduledTasks in memory until the server crashed.

After the 0.9 release, any more than 1,000 ScheduledTasks for a certain TaskDef are not loaded into memory but left on disk.

Principal Deletion​

The 0.9 release includes the ability to delete a Principal. The rpc DeletePrincipal is smart enough to ensure that there is always at least one Admin Principal to prevent a user from locking themselves out of the cluster.

PollThread in Java Task Worker​

We refactored the internal implementation of the Java Task Worker so that, for each LH Server in the cluster, the Task Worker creates a single PollThread object which is responsible for polling and executing TaskRuns. The PollThreads now poll in parallel, drastically increasing the throughput of a single Java Task Worker.

The PollThread was introduced in #796.

What's Next​

Our wire protocol (the GRPC API) is quite stable; there have been no major breaking changes since we introduced the alpha version of Multi-Tenancy in 0.7. We are diligently proceeding through soak tests, load tests, and chaos tests with our server and we have found and addressed several issues.

We continue to look foward to the 1.0 release, and we will reach that milestone once:

  • We are satisfied with results of load tests and soak tests.
  • We have had language experts review each of our three main SDK's (Java, Go, Python) and we have addressed any change requests.
  • We approach a year without any breaking changes to our wire protocol.