@camstack/server 1.2.272 → 1.2.273
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
|
@@ -27,8 +27,25 @@ exports.deliverAddonToAgent = deliverAddonToAgent;
|
|
|
27
27
|
* 3. `$agent.reload` — re-instantiate the addons from the new dist, and
|
|
28
28
|
* report which ones.
|
|
29
29
|
*/
|
|
30
|
-
/**
|
|
31
|
-
|
|
30
|
+
/**
|
|
31
|
+
* `$agent.deploy` is the INSTALL, not a control call.
|
|
32
|
+
*
|
|
33
|
+
* This constant used to be 30 s and said "install work is bounded well inside
|
|
34
|
+
* this". On 2026-09-11 that claim cost `little-unraid` an evening of running a
|
|
35
|
+
* previous `addon-pipeline` dist: three deploys, three
|
|
36
|
+
* `Request is timed out when call '$agent.deploy'` at exactly 30 s, while the
|
|
37
|
+
* CLI printed `✓ Deployed to 1 target(s)` each time — true of the hub, silent
|
|
38
|
+
* about the agent. The Mac answered the same call in under a second every time;
|
|
39
|
+
* this node keeps `/data` on an shfs/FUSE mount, where unpacking the pipeline
|
|
40
|
+
* bundle is minutes (the same mount makes its server `npm install` take ~15).
|
|
41
|
+
*
|
|
42
|
+
* `AGENT_BACKFILL_RPC_TIMEOUT_MS` had already learned this and written it down
|
|
43
|
+
* for the SAME RPC. Two constants for one call drifted, and the drift is
|
|
44
|
+
* invisible from the CLI: a bundle that lands while the reload never runs looks
|
|
45
|
+
* exactly like a deploy that worked. `agent-deploy-timeout.spec.ts` pins the
|
|
46
|
+
* relationship so they cannot separate again.
|
|
47
|
+
*/
|
|
48
|
+
exports.AGENT_DEPLOY_CONTROL_TIMEOUT_MS = 300_000;
|
|
32
49
|
/**
|
|
33
50
|
* `$agent.reload` re-forks group runners and re-registers their capabilities —
|
|
34
51
|
* legitimately slow on a node with a Python inference pool.
|