nandi/glean · 863259c

⑂ 863259c

Update the Modal readme for scale-to-zero

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

nandithebull committed 2026-09-21T19:11:38-07:00 Browse files

863259c parent: 49c0781

modified deploy/modal/readme.md +7 -5 Viewed

@@ -44,9 +44,10 @@ running its exit handler, up to five minutes of writes are gone. Lower
44	44	`max_containers=1` is not tuning, it is a correctness requirement. Two
45	45	containers would mean two SQLite writers over two independent copies of the
46	46	data, and the Volume's last-write-wins semantics would silently discard one of
47		-them. `min_containers=1` keeps that container alive because Glean's background
48		-workers — the Jetstream websocket consumer and the PDS sync / clustering /
49		-feed-fetch loops — only run while a container exists.
	47	+them. `min_containers=0` lets that container scale to zero when idle. Glean's
	48	+background workers — the Jetstream websocket consumer and the PDS sync /
	49	+clustering / feed-fetch loops — only run while a container exists, so they
	50	+pause once it scales down and resume when the next request wakes one.
50	51
51	52	## Secrets
52	53
@@ -75,8 +76,9 @@ Modal's model is autoscaling ephemeral containers. Glean is a stateful,
75	76	single-writer, always-on server. The mismatch is real and this deployment
76	77	manages it rather than resolving it:
77	78
78		-- One container, always warm. No horizontal scaling; `min_containers=1`
79		- means you pay for an idle container rather than scaling to zero.
	79	+- One container, scaled to zero when idle. No horizontal scaling, and
	80	+ with `min_containers=0` the first request after an idle period pays a cold
	81	+ start while the background workers are stopped in between.
80	82	- Snapshot durability, not continuous durability. See the window above.
81	83	- Container churn costs data. Any restart that skips the exit handler — a
82	84	hard kill, an OOM, a preemption — loses the interval.

  `max_containers=1` is not tuning, it is a correctness requirement. Two
  containers would mean two SQLite writers over two independent copies of the
  data, and the Volume's last-write-wins semantics would silently discard one of
--them. `min_containers=1` keeps that container alive because Glean's background
++them. `min_containers=0` lets that container scale to zero when idle. Glean's
--workers — the Jetstream websocket consumer and the PDS sync / clustering /
++background workers — the Jetstream websocket consumer and the PDS sync /
--feed-fetch loops — only run while a container exists.
++clustering / feed-fetch loops — only run while a container exists, so they
++pause once it scales down and resume when the next request wakes one.
  ## Secrets
  single-writer, always-on server. The mismatch is real and this deployment
  manages it rather than resolving it:
--- **One container, always warm.** No horizontal scaling; `min_containers=1`
++- **One container, scaled to zero when idle.** No horizontal scaling, and
--  means you pay for an idle container rather than scaling to zero.
++  with `min_containers=0` the first request after an idle period pays a cold
++  start while the background workers are stopped in between.
  - **Snapshot durability, not continuous durability.** See the window above.
  - **Container churn costs data.** Any restart that skips the exit handler — a
    hard kill, an OOM, a preemption — loses the interval.