feat: support multiple rate limits #11260
Conversation
xl-openai
force-pushed
the
xl/skill-limit
branch
3 times, most recently
from
f0abfb0 to
5a1de97
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5a1de97fb3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| .find(|snapshot| snapshot.limit_name.as_deref() == Some("codex")) | ||
| .cloned() | ||
| .unwrap_or_else(|| snapshots[0].clone()); |
There was a problem hiding this comment.
This selection always prefers the "codex" entry, but get_rate_limits_many now always includes a synthetic codex snapshot even when /usage returns rate_limit: null and only non-codex buckets have real window data. In that case rate_limits in GetAccountRateLimitsResponse is returned empty while rate_limits_by_name has populated limits, which regresses legacy clients that still read only rate_limits. Prefer codex only when it actually contains window/credit data, otherwise fall back to a populated snapshot.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
rate_limits is intentionally the canonical codex bucket for legacy clients; falling back to additional buckets would change its semantics and misrepresent what legacy clients are actually able to interpret.
| HeaderValue::from_static("1704074400"), | ||
| ); | ||
|
|
||
| let snapshot = parse_rate_limit_for_limit(&headers, Some("codex_other")).expect("snapshot"); |
There was a problem hiding this comment.
This test sets x-codex-secondary-* headers but calls parse_rate_limit_for_limit(..., Some("codex_other")), so the parser looks for x-codex-other-* keys and cannot populate primary, making the subsequent expect("primary") fail. The limit name and header prefix need to match (for example codex_secondary with x-codex-secondary-*).
Useful? React with 👍 / 👎.
xl-openai
force-pushed
the
xl/skill-limit
branch
6 times, most recently
from
9f74d29 to
dd06980
Compare
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dd069802d6
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| self.rate_limit_snapshots_by_limit_id | ||
| .insert(limit_id, display); |
There was a problem hiding this comment.
This path only upserts a single bucket into rate_limit_snapshots_by_limit_id, and buckets are cleared only when snapshot is None. With the new multi-limit polling flow, snapshots are emitted one-by-one from /usage, so if a bucket disappears from a later response (for example, codex_other is no longer returned), there is no tombstone and the old entry remains in memory, causing /status to keep rendering stale limits indefinitely.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
I think it is OK for now.
| let suffix = "-primary-used-percent"; | ||
| let prefix = header_name.strip_suffix(suffix)?; |
There was a problem hiding this comment.
parse_rate_limit_updates discovers additional buckets via header_name_to_limit_id, but this helper only matches headers ending in -primary-used-percent. If a non-default bucket is reported with only secondary-window headers, that bucket is never discovered, so no snapshot is emitted and consumers miss the update entirely. Limit-id discovery should include secondary header families (or use a broader prefix-based detection) to avoid dropping valid buckets.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
This should no happen.
| limit_id: None, | ||
| limit_name: None, |
There was a problem hiding this comment.
are the core rate limits name/idless?
There was a problem hiding this comment.
ok i see you store them rate_limits
There was a problem hiding this comment.
I can update the test with a real value of limit_id/name.
There was a problem hiding this comment.
do we still update v1 schemas?
There was a problem hiding this comment.
Must be auto-generated from some command I run. Let me revert.
There was a problem hiding this comment.
just write-app-server-schema actually it will still generate v1 (otherwise C1 will fail).
xl-openai
force-pushed
the
xl/skill-limit
branch
from
dd06980 to
7c81f7e
Compare
codex-rs/core/src/error.rs
Outdated
| impl std::fmt::Display for UsageLimitReachedError { | ||
| fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { | ||
| // Pro users might hit a non-standard codex metered bucket. | ||
| if matches!(self.plan_type, Some(PlanType::Known(KnownPlan::Pro))) |
There was a problem hiding this comment.
does it have to gate to pro or can it be more general?
xl-openai
force-pushed
the
xl/skill-limit
branch
from
7c81f7e to
b494b5b
Compare
3 participants
Added multi-limit support end-to-end by carrying limit_name in rate-limit snapshots and handling multiple buckets instead of only codex.
Extended /usage client parsing to consume additional_rate_limits
Updated TUI /status and in-memory state to store/render per-limit snapshots
Extended app-server rate-limit read response: kept rate_limits and added rate_limits_by_name.
Adjusted usage-limit error messaging for non-default codex limit buckets