Saner defaults for configs

[gouthamve]

Signed-off-by: Goutham Veeramachaneni [email protected]

What this PR does:

Which issue(s) this PR fixes:
Fixes #

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[bboreham]

Seems reasonable, but I don't understand why you'd want so many tokens.
Also, what happens if you raise the number without noticing? Say you had 4 old ingesters with 128 tokens and you add a new one with 512 - does it get half the series?

[bboreham]

Should update the docs to remove advice to set these, or at least note it's only for older versions.

Would we take these out of the example manifests?

[tomwilkie]

Seems reasonable, but I don't understand why you'd want so many tokens.

More tokens == better load balancing - less variation is the size of the ranges owned by the ingesters.

Also, what happens if you raise the number without noticing? Say you had 4 old ingesters with 128 tokens and you add a new one with 512 - does it get half the series?

More tokens will mean more ranges, but those ranges will be smaller, so its not quite linear IIRC.

[bboreham]

That’s a fine theory, but do we have any evidence? Has anyone ever noticed an imbalance (except me, running with 512 tokens per ingester)?

I think giving one new ingester a large fraction of the series is likely to be disastrous. Suggest this particular change should be left in the documentation.

[pracucci]

LGTM. I left few comments, but overall looks good.

[bboreham]

Can't set spread-flushes and leave jitter non-zero.

[bboreham]

Do we want to remove recommendations from the docs for what is now the default?
E.g. https://github.com/cortexproject/cortex/blame/b3112b3315da58eb0ce3b50caa8df05f0a633db8/docs/guides/running.md#L244-L248

[bboreham]

lgtm