[fully_async] fix: fully_async ckpt save bug by sl-1314 · Pull Request #5677 · verl-project/verl

sl-1314 · 2026-03-20T08:11:04Z

What does this PR do?

fix fully_async ckpt save bug

Checklist Before Starting

Search for similar PRs. Paste at least one query link here: ...
Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
- {modules} include fsdp, megatron, veomni, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data, cfg, reward, fully_async, one_step_off
- If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
- {type} is in feat, fix, refactor, chore, test
- If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
- Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc.

API and Usage Example

Demonstrate how the API changes if any, and provide usage example(s) if possible.

# Add code snippet or script demonstrating how to use this

Design & Code Changes

Demonstrate the high-level design if this PR is complex, and list the specific changes.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

Read the Contribute Guide.
Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
Add / Update the documentation.
Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)
If your PR is related to the recipe submodule, please also update the reference to the submodule commit via git submodule update --remote or cd recipe && git pull origin main.

gemini-code-assist

Code Review

This pull request fixes a bug in the checkpoint saving logic for fully asynchronous training. The change from and to or correctly ensures that a forced save is not dependent on the save frequency step. My review includes a suggestion to further refactor the condition to make it more robust, by decoupling forced saves from the save_freq setting, which would prevent accidental data loss if periodic saving is disabled.

gemini-code-assist · 2026-03-20T08:15:00Z

        if self.config.trainer.save_freq > 0 and (
-            force and self.current_param_version % self.config.trainer.save_freq == 0 or esi_close_to_expiration
+            force or self.current_param_version % self.config.trainer.save_freq == 0 or esi_close_to_expiration
        ):


While this change correctly fixes the logical operator precedence, the overall condition for saving a checkpoint could be made more robust and less redundant.

Forced saves are dependent on save_freq: The entire condition is gated by self.config.trainer.save_freq > 0. This means if a user sets save_freq <= 0 to disable periodic saves, it will also prevent forced saves (e.g., on the last step or when an ESI instance is expiring). This could lead to unexpected loss of progress.

Redundant esi_close_to_expiration check: The esi_close_to_expiration check is performed both at the call site (and its result is passed via the force flag) and again inside this function.

To address these points, I suggest refactoring the condition to decouple periodic saves from forced saves. This also makes the logic clearer.

Suggested change

if self.config.trainer.save_freq > 0 and (

force and self.current_param_version % self.config.trainer.save_freq == 0 or esi_close_to_expiration

force or self.current_param_version % self.config.trainer.save_freq == 0 or esi_close_to_expiration

):

if (

force

or esi_close_to_expiration

or (self.config.trainer.save_freq > 0 and self.current_param_version % self.config.trainer.save_freq == 0)

):

### What does this PR do? fix fully_async ckpt save bug ### Checklist Before Starting - [ ] Search for similar PRs. Paste at least one query link here: ... - [ ] Format the PR title as `[{modules}] {type}: {description}` (This will be checked by the CI) - `{modules}` include `fsdp`, `megatron`, `veomni`, `sglang`, `vllm`, `rollout`, `trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`, `ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`, `env`, `tool`, `ckpt`, `doc`, `data`, `cfg`, `reward`, `fully_async`, `one_step_off` - If this PR involves multiple modules, separate them with `,` like `[megatron, fsdp, doc]` - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test` - If this PR breaks any API (CLI arguments, config, function signature, etc.), add `[BREAKING]` to the beginning of the title. - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching` ### Test > For changes that can not be tested by CI (e.g., algorithm implementation, new model support), validate by experiment(s) and show results like training curve plots, evaluation results, etc. ### API and Usage Example > Demonstrate how the API changes if any, and provide usage example(s) if possible. ```python # Add code snippet or script demonstrating how to use this ``` ### Design & Code Changes > Demonstrate the high-level design if this PR is complex, and list the specific changes. ### Checklist Before Submitting > [!IMPORTANT] > Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review. - [ ] Read the [Contribute Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md). - [ ] Apply [pre-commit checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting): `pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always` - [ ] Add / Update [the documentation](https://github.com/volcengine/verl/tree/main/docs). - [ ] Add unit or end-to-end test(s) to [the CI workflow](https://github.com/volcengine/verl/tree/main/.github/workflows) to cover all the code. If not feasible, explain why: ... - [ ] Once your PR is ready for CI, send a message in [the `ci-request` channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the `verl` Slack workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ). (If not accessible, please try [the Feishu group (飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).) - [ ] If your PR is related to the `recipe` submodule, please also update the reference to the submodule commit via `git submodule update --remote` or `cd recipe && git pull origin main`.

fix: fully_async ckpt save bug

9b68d87

sl-1314 requested review from ArronHZG and wuxibin89 as code owners March 20, 2026 08:11

ArronHZG enabled auto-merge (squash) March 20, 2026 08:12

ArronHZG approved these changes Mar 20, 2026

View reviewed changes

ArronHZG merged commit 8e24127 into verl-project:main Mar 20, 2026
50 of 62 checks passed

gemini-code-assist Bot reviewed Mar 20, 2026

View reviewed changes

Navigation Menu

Repository navigation

[fully_async] fix: fully_async ckpt save bug - #5677

[fully_async] fix: fully_async ckpt save bug#5677
ArronHZG merged 1 commit into
verl-project:mainfrom
meituan-search:fix_fully_async0

sl-1314 commented Mar 20, 2026

Uh oh!

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Mar 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

sl-1314 commented Mar 20, 2026

What does this PR do?

Checklist Before Starting

Test

API and Usage Example

Design & Code Changes

Checklist Before Submitting

Uh oh!

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Mar 20, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants