only fetches one chunk per series in /series

[owen-d]

What

Improves /series API by fetching a single chunk for each series and then appying matchers one time per series.

Why

Previously we used used the same functions and associated EntryIterator ifc as the traditional query path, fetching all logs in a time range then calculating labels per entry, only deduping them at the end. This was wildly inefficient and would oom queriers.

Proof

Tested locally w/ v9 schema

$ ls -lh /var/log/*.log | grep -v ' 0B ' | wc -l
       9
$ logcli series --match='{filename=~".*log$"}' | wc -l
http://localhost:3100/loki/api/v1/series?end=1586288927776455000&match=%7Bfilename%3D~%22.%2Alog%24%22%7D&start=1586285327776455000
       9

/cc @cyriltovena

[cyriltovena]

LGTM, way to go !