Re: [pcp] RFC: allowing longer metric and instance names in MMV(5) format

[Suyash <dextrous93@xxxxxxxxx>]

Hi everyone,

Suyash from speed here

Sorry for the late response.

I have implemented the strings block for speed yesterday, and updated the simple example in the repo to reflect that. Currently the written mmv file is not visible to pmdammv, but will hopefully get that done by this week's end.

I completely agree with the idea of backwards compatibility, and from speed's standpoint, it is only about turning the metric name to `*PCPString` type instead of a normal `string` type and caching the address where it is written.

The address can be calculated dynamically, at the time of registering a metric. We can check the length and if it is more than 63, we can add it to the string block, otherwise use the v1 method. The address is of 8 bytes, and we have 63 to work with, so we can add a flag (lets say zero padding) for the first 8 bytes, to tell pmdammv that the name for this metric is in the strings block, and the next 8 bytes have the offset. Writing can be modified by detecting the offset at the time of writing the metric desc. We can do a similar scheme for instances and instance domains also.

The only concern is we will still use 64 bytes for metric name for each metric, despite needing only need 8, but we use 64 bytes anyway, no matter what the length of the actual name (https://github.com/performancecopilot/speed/blob/master/writer.go#L302).

- Suyash