Skip to content

Extract the Ray actor construction into launch_sglang_ray_actor - #1868

Open
fzyzcjy wants to merge 1 commit into
tom/refactor-miles/op5-4from
tom/refactor-miles/op5-5
Open

Extract the Ray actor construction into launch_sglang_ray_actor#1868
fzyzcjy wants to merge 1 commit into
tom/refactor-miles/op5-4from
tom/refactor-miles/op5-5

Conversation

@fzyzcjy

@fzyzcjy fzyzcjy commented Jul 28, 2026

Copy link
Copy Markdown
Collaborator

Part of #1837

@gemini-code-assist

Copy link
Copy Markdown
Contributor

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from 40b61fb to 7f95ba9 Compare July 28, 2026 13:31
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 03c7b85 to 0f6ba81 Compare July 28, 2026 13:31
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from 7f95ba9 to b28729e Compare August 3, 2026 23:45
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 0f6ba81 to 805ddc7 Compare August 3, 2026 23:45
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from b28729e to 9e31799 Compare August 4, 2026 00:39
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 805ddc7 to 328201e Compare August 4, 2026 00:39
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from 9e31799 to 7856256 Compare August 4, 2026 06:58
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 328201e to caa6f9f Compare August 4, 2026 06:58
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from 7856256 to 9b2b0dd Compare August 6, 2026 07:03
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from caa6f9f to 072d6e3 Compare August 6, 2026 07:03
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from 9b2b0dd to c334e14 Compare August 10, 2026 06:07
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch 2 times, most recently from ad0b248 to cba62b2 Compare August 10, 2026 16:07
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from c334e14 to 95fdd1e Compare August 10, 2026 16:07
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from cba62b2 to ac9ee52 Compare August 10, 2026 16:34
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch 2 times, most recently from 4c50796 to 8240808 Compare August 12, 2026 02:48
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch 3 times, most recently from be4bfaa to da97df0 Compare August 12, 2026 23:56
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from b3dc110 to 842a79f Compare August 13, 2026 01:13
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from da97df0 to 7c49565 Compare August 13, 2026 01:13
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from 842a79f to b9a2ee9 Compare August 13, 2026 03:25
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 7c49565 to 2dd13f8 Compare August 13, 2026 03:25
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-4 branch from b9a2ee9 to 55146af Compare August 13, 2026 06:01
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 2dd13f8 to 3cd8662 Compare August 13, 2026 06:01
The free function lives in server_cell.py so later ops can move engine
startup onto the cell; start_engines now only decides which slots to fill.
Test fixtures patch SGLangEngine at its new home.
@fzyzcjy
fzyzcjy force-pushed the tom/refactor-miles/op5-5 branch from 3cd8662 to 80bf524 Compare August 14, 2026 04:46
guapisolo
guapisolo previously approved these changes Aug 24, 2026

@guapisolo guapisolo left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed in the #1837 stack context: this is a behavior-preserving extraction of the SGLang Ray actor construction, with placement, environment, and lifecycle flow preserved. No active finding needs to be carried into safe merge checkpoint 1 (#2176).

) -> ray.actor.ActorHandle:
pg, reordered_bundle_indices, reordered_gpu_ids = pg

num_gpus = 0.2

@guapisolo guapisolo Aug 25, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[P1] Ban the RDT path before landing this refactor

RDT is not part of this refactor plan and conflicts directly with the actor-ownership changes in this stack. Do not port the #1313 launch logic into launch_sglang_ray_actor.

Minimal revert instruction:

  1. Remove "rdt" from the --update-weight-transfer-mode choices so startup rejects this path.
  2. Resolve this hunk entirely in favor of this PR: drop rdt_reuse_pg, the RDT num_gpus=0 branch, rdt_pg_kwargs, and pg_bundles forwarding.
  3. Keep launch_sglang_ray_actor as the single non-RDT actor-launch path; do not add an RDT compatibility branch.

The remaining unreachable RDT implementation can be removed separately. The acceptance condition here is that --update-weight-transfer-mode rdt is rejected and no RDT-specific rollout-launch behavior survives this rebase.

@guapisolo
guapisolo dismissed their stale review August 25, 2026 01:54

need action

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants