Preserve upstream turn_rate_limit_exceeded errors and retry advice without
cooling the account or replaying the request inside the proxy.
Cover HTTP and WebSocket rejections plus streamed failures, and document
client-owned waiting and retries.
Co-authored-by: Codex <noreply@openai.com>
Distinguish credential revision conflicts from account store failures and
persist response cookies once across streaming errors and WebSocket reuse.
Keep Microsoft Store downloads first with a 30-second timeout, read official
fallback sizes through bounded HEAD requests, and correct Store size units.
Co-authored-by: Codex <noreply@openai.com>
Default batch scheduling to disabled and always submit the displayed
switch value, allowing an untouched form to apply that setting.
Keep sidebar focus rings inside navigation buttons to prevent clipping
without changing their layout or keyboard behavior.
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Accept positive fractional days and expire request dump segments by elapsed time. Rotate small dump files by age, preserve archive timestamps, and document the retention window.
Co-authored-by: Codex <noreply@openai.com>
Scheduled account warmup sent reasoning.effort=none, but the Codex model
catalog lists low as the lowest supported level for every model an account
can see. Upstream answers HTTP 200 with an SSE error event (unsupported_value
on reasoning.effort), which the warmup terminal check records as
response_failed, so the cycle ends with warmed_up=0 and never starts the 5h
window.
Send low instead and cover the upstream payload with a regression test.
Deliver HTTP/SSE and WebSocket events immediately while preserving bounded
recovery after replaceable preludes. Share prelude classification so response
metadata does not hide WebSocket connection-limit recovery.
Keep irreversible delivery, cancellation, middleware, and retry-budget guards.
Document the distinction between transport keepalive and event idle timeouts.
Refs #405
Co-authored-by: Codex <noreply@openai.com>
Merge migrations 0030 through 0032 after the v3.25.0 release baseline. Queue analytics bucket rebuilds once and verify upgrade preservation and repeated startup.
Co-authored-by: Codex <noreply@openai.com>
Aggregate analytics in PostgreSQL, reconcile synchronous usage with pending source changes, and preserve exact pagination and forecast totals across large histories.
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Stage scoped body fragments and reuse context locations while preserving per-rule rollback. Account for encoded size incrementally and use bounded work plus cancellation instead of a fixed wall-clock cutoff.
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Color first-output and total latency independently using fixed thresholds.
Keep the existing layout and details popover without a side indicator.
Closes#407
Co-authored-by: Codex <noreply@openai.com>
* fix(accounts): preserve quota forecast totals on same-window reversals
Share exhausted-window recovery evidence between account recovery and
forecast sampling. Keep recorded cycle usage on ordinary reversals and
restart totals only after the recovery evidence is satisfied.
Co-authored-by: Codex <noreply@openai.com>
* fix(accounts): separate forecast cycle boundaries from quota recovery
Keep same-window usage totals after quota reversals, including observations near exhaustion. Leave access recovery in the OpenAI provider and reset forecast totals only at an identified window or plan boundary.
Co-authored-by: Codex <noreply@openai.com>
---------
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Let Core decide replay eligibility after WebSocket preludes or HTTP
keepalives. Persist subsequent attempts while still enforcing the running
lifecycle and consecutive attempt numbers, preserving delivery facts and
the upstream send watermark.
Cover retry metadata, duplicate and skipped attempts, and terminal request
immutability with a PostgreSQL regression test.
Co-authored-by: Codex <noreply@openai.com>
Start downstream keepalives while admitted streaming requests await their first output, including middleware processing. Separate transport commitment from the replay barrier so heartbeat frames preserve safe retries and execution accounting.
Co-authored-by: Codex <noreply@openai.com>
Include incomplete final outcomes in service health metrics and lower the
stable threshold to 95%. Keep the health popover at four outcomes by
folding incomplete requests into the service failure display.
Use consistent service success rates across dashboard trends, health
timelines and usage insights. Cover the reported counts, threshold edges,
caller exclusions and final success after failed attempts in regressions.
Refs #406
Co-authored-by: Codex <noreply@openai.com>
Add published GPT-6 Astra and GPT-6.1 Sol rates for short and long context. Expose Ultrafast override bands in the pricing API and editor, with billing and validation coverage.
Co-authored-by: Codex <noreply@openai.com>
The analytics worker acknowledged projected queue versions inside its
repeatable-read projection transaction. Request updates keep rewriting
pending queue rows through the source triggers, and once such a change
commits after the worker snapshot, the acknowledgement row lock fails with
a serialization error instead of skipping the stale version. Under steady
traffic every cycle rolled back the whole projection batch, the queue kept
growing and the worker stayed in backoff.
Commit the projection and publication first, then acknowledge the exact
versions in a separate read-committed transaction. Row locks there
re-check the latest committed version, so rewritten versions no longer
match and stay queued for replay, while locked rows are still skipped.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Preserve known model ownership when provider catalog refreshes fail.
Reject ambiguous unknown-provider routes and exclude ChatGPT OAuth
accounts from Grok requests while preserving API key passthrough.
Refs #391
Co-authored-by: Codex <noreply@openai.com>
Successful responses keep refreshing the quota observation time of the
account they used. The post-selection credential reload compared the
whole account snapshot, so a busy account could hit three observation-only
conflicts in a row and fail with account_snapshot_conflict before any
upstream request was sent.
Accept the reloaded snapshot when only the observation time of a
non-exhausted quota moved. Credential, state, scheduling, scope and quota
conclusion changes still invalidate the selection and retry as before.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Pause Runtime and SDK processing budgets while middleware delegates to host next, then resume the remaining budget. Preserve request deadlines and cancellation, share callback deadline state, and add timeout diagnostics and regression coverage.
Fixes#392
Co-authored-by: Codex <noreply@openai.com>
Hold PostgreSQL observation slots through connection cleanup and optimize exact usage pagination. Keep usage panels visible when analytics queries fail and backfill missing historical identity snapshots in one migration.
Co-authored-by: Codex <noreply@openai.com>
Fold migrations 0028 through 0035 into one migration while preserving every released migration through v3.23.1. Define the final observation validators, history schema, recovery leases, token columns, indexes, and analytics triggers once.
Refresh the migration manifest, remove superseded intermediate upgrade tests, and retain final-schema business validation. Verify clean installation, released-baseline upgrades, schema equivalence, and historical data using temporary PostgreSQL checks.
Co-authored-by: Codex <noreply@openai.com>
Prefer providers whose catalogs contain the mapped model before providers accepting unlisted models. Preserve account scope, route constraints, provider order within each tier, and unknown-model passthrough.
Add compiled mixed-catalog regressions and document candidate ordering. Refs #391.
Co-authored-by: Codex <noreply@openai.com>
* fix(openai): preserve Lite interrupts across prelude retries
Deliver known Responses Lite WebSocket preludes before semantic output while retaining existing recovery guards. Keep interrupt intent across retries and retarget it only after the replacement response is delivered.
Co-authored-by: Codex <noreply@openai.com>
* fix(openai): forward Lite events without waiting for output
Deliver unclassified Lite WebSocket events in order and reuse the existing commit boundary to stop replay after their delivery. Preserve error classification, retry budgets, account recovery, and interrupt handling.
Co-authored-by: Codex <noreply@openai.com>
---------
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
Build durable request facts and bounded aggregate queries, freeze historical
identities, and move recovery lease updates out of request history rows.
Align token fields across storage, APIs and pages.
Invalidate analytics only after actual request deletions and track embedded
migration inputs during builds.
Co-authored-by: Codex <noreply@openai.com>
Record observed closure before terminal delivery can block, reject queued
unsent commands, and skip commands cancelled before socket writes.
Preserve connection diagnostics and test account egress across quota writes.
Co-authored-by: Zyy <zyycn@users.noreply.github.com>
Co-authored-by: Codex <noreply@openai.com>
ZyyandCodex
dudukee
JiahaoandCodex
think-backandClaude Opus 5.5
dependabot[bot]
hi_yuebanandCodex
