Add 0.18.0rc1 changelog #1535
Add 0.18.0rc1 changelog #1535
Conversation
761fd86 to
548c915
Compare
| * [FEATURE] Add `holt_winters` query function | ||
| * [FEATURE] Add relative complement `unless` operator to PromQL | ||
| * [ENHANCEMENT] Handle data corruption more gracefully | ||
| * [ENHANCEMENT] Quarantine series file on write issues |
There was a problem hiding this comment.
I think these two items are essentially the same (or the latter is the way to do the former).
There was a problem hiding this comment.
Perhaps: "Quarantine series file if data corruption is encountered (instead of crashing)."
There was a problem hiding this comment.
Or "instead of panic" if we want to use the Go lingo.
|
I assume we will add same more instructions to the announcement mail (like how to activate the varbit encoding and such). |
|
Yes. What's "and such"? I don't see anything else configuration wise – PromQL changes are found in the documentation, which we can link. We didn't do this before though. |
|
👍 |
|
"and such" could be the explanation about the improvement in query time if you have a lot of series churn (so that people can do the trade off if they want to upgrade or not - for those in less agile environment where an upgrade is a big deal). |
|
Can you give me a short paragraph to include here? |
548c915 to
6d1a756
Compare
|
Please publish a blog post about the release and all the necessary information (explanations, links, howtos, etc.). The mailing list is nice but not widely used. |
|
I'll phrase a paragraph about the chunk encoding and the index lookups once I'm out of the meeting I'm currently in |
|
@fabxc Suggestion: This release introduces a new chunk encoding with variable bit width, in parts inspired by Facebook's "Gorilla" paper. At the price of increased CPU cost and run-time of queries, the chunk space requirement drops by ~60% for typical data sets. The varbit encoding has to be enabled with the command line flag Several storage improvements dramatically increase the performance of instant queries for the most recent value of time series, as it typically happens in rule evaluations and for federation. On servers with a lot of series churn (i.e. many older series that don't receive samples anymore), we have seen a 12x speedup of rule evaluations with CPU usage reduced by 75%. |
|
Great numbers. Really nice work! |
@brian-brazil