codeintel: Replace SQLite with dgraph-io/badger

[efritz]

No description provided.

[codecov]

Codecov Report

Merging #11052 into tar-bundle-payloads will not change coverage.
The diff coverage is n/a.

@@                 Coverage Diff                  @@
##           tar-bundle-payloads   #11052   +/-   ##
====================================================
  Coverage                46.34%   46.34%           
====================================================
  Files                     1374     1374           
  Lines                    77412    77412           
  Branches                  6595     6595           
====================================================
  Hits                     35875    35875           
  Misses                   38130    38130           
  Partials                  3407     3407           

Flag	Coverage Δ
#go	50.38% <0.00%> (ø)
#storybook	6.88% <0.00%> (ø)
#typescript	35.84% <0.00%> (ø)
#unit	46.13% <0.00%> (ø)

[unknwon]

Is there an RFC for this change?

The default memory footprint of badger is not small. Also it requires 64-bits system to compile (even though they claimed fixed this many times but I couldn't compile it on a 32-bit machine as of Jan or Feb this year).

[efritz]

Is there an RFC for this change?

No, this is just an experiment. I plan to do measurements with a handful of different storage mechanisms and serialization techniques.

The default memory footprint of badger is not small. Also it requires 64-bits system to compile (even though they claimed fixed this many times but I couldn't compile it on a 32-bit machine as of Jan or Feb this year).

I'll do additional profiling to see if I can get it down. Is it necessary to compile on a 32-bit machine?

[keegancsmith]

@efritz I think sqlite is just really good at immutable data stores? When you benchmarked the performance, where did the issues lie? Is it possible it is just that we are always dominated by serialization so it doesn't really matter what the underlying store is?