Make local dev environment use ~650MB less memory

[cdrini]

With AI, it's more common that we might want to run multiple instances of the local dev environment at once, and memory usage is becoming a bottleneck for that workflow. Investigated with Claude, and found some opportunities to reduce memory usage:

• Since the number of documents solr is very small, we can reduce the solr heap size from the default of 512 MB to have the local environment use less memory. That wins ~400MB
• We no longer need multiple web workers; this was once required for login to work. Reducing to 1 web workers gives us another ~250MB.

That sends the local dev environment from using 1.6GB of memory down to about 1GB of memory! Still feels too high for open library, but it's an improvement.

Technical

Here's the overall memory usage of each container -- note this is after reducing solr's heap size.

xychart-beta
    title "Per-service memory (MB)"
    x-axis "Elapsed (s)" [0, 7, 15, 22, 29, 37, 44, 51, 58, 66, 73, 80, 88, 95, 102, 109, 116]
    y-axis "Memory (MB)" 0 --> 500
    line [428 "solr", 426, 426, 426, 428, 432, 431, 436, 452, 456, 453, 456, 455, 458, 456, 455, 455 "solr"]
    line [395 "web", 396, 396, 396, 393, 393, 392, 392, 391, 391, 391, 392, 392, 390, 392, 392, 392 "web"]
    line [219 "fast_web", 271, 271, 271, 272, 274, 274, 273, 273, 274, 276, 275, 275, 275, 275, 275, 275 "fast_web"]
    line [9 "solr-updater", 7, 6, 7, 120, 260, 228, 224, 223, 224, 223, 224, 224, 224, 224, 224, 224 "solr-updater"]
    line [10 "home", 6, 6, 7, 7, 48, 150, 144, 143, 32, 252, 0 "home exits", 0, 0, 0, 0, 0 "home"]
    line [100 "covers", 100, 100, 100, 100, 101, 101, 105, 106, 112, 111, 112, 112, 115, 113, 113, 113 "covers"]
    line [53 "db", 56, 62, 58, 60, 66, 75, 71, 71, 69, 71, 66, 67, 67, 71, 67, 67 "db"]
    line [53 "infobase", 53, 53, 53, 53, 53, 54, 54, 54, 54, 54, 54, 54, 54, 54, 54, 54 "infobase"]
    line [49 "mockservices", 49, 49, 49, 50, 53, 49, 50, 50, 50, 50, 50, 54, 50, 50, 50, 50 "mockservices"]
    line [7 "memcached", 7, 7, 7, 7, 7, 18, 18, 18, 19, 19, 18, 18, 18, 18, 18, 18 "memcached"]

Loading

solr settles highest (~455MB even with the 128MB heap — JVM overhead beyond the heap itself), followed by
web (~392MB, flat throughout — already warm) and
fast_web (~275MB, quick ramp at startup).
solr-updater ramps to ~224MB once it connects and starts polling.
home (the reindex process) ramps up, peaks ~252MB, then drops to 0 when it exits around 80s. The rest stay low and flat: covers (~113MB),
db (~67MB), infobase (~54MB), mockservices (~50MB),
memcached (~18MB).

Testing

Solr heap

Confirmed editing works, searching works, facets work.

• Baseline (default Solr heap, -Xms512m -Xmx512m): home container exited (reindex complete) after 67.9s wall time.
• SOLR_HEAP=128m: home container exited after 75.0s wall time.

The ~7s difference is within the noise of this single run each (network variance from the docs-wiki clone home does on startup, etc.) — heap size had no meaningful effect on how long the reindex itself took.

Samples were taken every ~7s (5s requested interval + docker stats overhead):

xychart-beta
    title "Solr container memory (MB)"
    x-axis "Elapsed (s)" [0, 7, 15, 22, 29, 36, 44, 51, 58, 66]
    y-axis "Memory (MB)" 0 --> 900
    line [798 "512MB Heap", 791, 791, 791, 792, 799, 824, 826, 830, 830 ]
    line [362 "128MB Heap", 362, 361, 362, 362, 364, 404, 403, 410, 413]

Loading

Web workers

Confirmed logging in works

xychart-beta
    title "web container memory (MB): 4 vs 2 vs 1 workers"
    x-axis "Sample" [1, 2, 3, 4, 5]
    y-axis "Memory (MB)" 0 --> 450
    line [393 "4 workers", 393, 391, 391, 390]
    line [224 "2 workers", 224, 224, 224, 223]
    line [134 "1 worker", 135, 135, 135, 135]

Loading

Screenshot

Stakeholders