You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For lexical, the index was using 450MB of disk and consumed about a
gig in memory. This commit includes the following changes:
* We stopped duplicating entries in a couple of the records, and now
only store them once in the 'by_ref' key. This decreased the index
size by 50%, and had no noticable effect on performance.
* Several of the processes had cache entries lying around that
caused them to consume quite a bit (hundreds of megs) of memory. I
introduced a cache cleaning block and added it to the call sites,
and this reduced the memory usage of those processes back to
extremely low levels.
* We were storing the erlang and elixir versions on each entry. Now we
create separate indexes for each elixir/erlang version pair
In total, lexical was taking around 1gb of memory before these
changes, and now it takes around 400mb. Disk space dropped from 450 megs
to 250.
0 commit comments