fix(rollout): make engine port range start configurable via MILES_ROLLOUT_START_PORT#64
Open
zhihengy wants to merge 1 commit into
Open
fix(rollout): make engine port range start configurable via MILES_ROLLOUT_START_PORT#64zhihengy wants to merge 1 commit into
zhihengy wants to merge 1 commit into
Conversation
zhihengy
added a commit
that referenced
this pull request
Jul 24, 2026
Host-level ops knob, orthogonal to the cosmos3 adaptation. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…LOUT_START_PORT The engine port scan starts at a hardcoded 15000, and the free-port probe misses servers bound to a specific interface. Two miles runs on one host therefore claim the same port range, and the second run's weight sync lands on the first run's engines (surfacing as "Invalid device_uuid"). Shifting the scan start per run keeps the ranges disjoint. Default is unchanged. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
zhihengy
force-pushed
the
fix/rollout-start-port
branch
from
July 24, 2026 22:26
4e46961 to
164429e
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Engine port allocation scans upward from a hardcoded
15000, and the free-port probe misses servers bound to a specific interface (rather than0.0.0.0). When two miles runs share a host, both claim the same port range: the second run's trainer connects to the first run's sglang engines and sends weight sync there, surfacing as an unrelated-lookingInvalid device_uuiderror mid-training.Fix
Make the scan start configurable via
MILES_ROLLOUT_START_PORTso concurrent runs on one host can use disjoint ranges (e.g.export MILES_ROLLOUT_START_PORT=16000for the second run). Default stays15000— behavior is unchanged unless the variable is set.Split out of the cosmos3 branch: this is host-level ops, orthogonal to any model adaptation.
🤖 Generated with Claude Code