Speed up filter processing #476
Conversation
|
Holy! This has made such an improvement to my dashboard! I was at the point of sacking off all the auto-entities i'd written across mutliple dashboards & replace them with fixed lists again. This has made a massive difference to responsiveness, especially on mobile devices with lower compute. @thomasloven please consider this MR. Until this is rolled in & released, I will be using the forked version. |
|
FWIW, I eventually went to writing my own dashboard strategy... not because of shortcomings here as much as the pain of trying to shove everything specific I wanted to do into nothing but templates & yaml. (Also because working on this PR taught me enough to make me more confident in writing my own HA-flavored frontend JS.) Anyhow, if you find any issues with my PR please LMK and I'll try to fix them up! |
|
This looks very interesting. Could you try to clean those out? |
|
Huh. Not intentional, will try to fix that. |
mrnerdhair
force-pushed
the
improve-filter-performance
branch
from
83ee40c to
67b79b5
Compare
|
Whoops! Looks like I accidentally reset my VS Code formatting defaults to Deno a while back. Here's a better breakdown. I've pinned a specific version of prettier (as they recommend) and normalized the formatting to that. It's still a pretty big changeset, but at least it's more readable now. |
| babel({ | ||
| exclude: "node_modules/**", | ||
| babelHelpers: "bundled", |
There was a problem hiding this comment.
This was the previous behavior; specifying it explicitly prevents a build-time warning.
|
@mrnerdhair tested out your branch on HA locally, performance looks a lot better! FWIW -- testing steps -- I had to create a new build of the hass-custom-devcontainer, forked here: https://github.com/bobbravo2/hass-custom-devcontainer That and the Once that was done, I was able to test your branch: |
Fixes #379.
Filtering and sorting didn't scale well; area and device processing in particular was O(n^2). In my environment, with ~3000 entities and ~200 devices, displaying a handful of rooms involved 30-second load times.
I've approached the problem by breaking the filter and sort functions down into sections to allow re-use of various intermediate results. I've also added an improved caching infrastructure to prevent multiple loads of the areas/devices/entities/labels lists being triggered by a flurry of initial cache misses.
These changes bring my ~30sec case down to ~200ms.