deps(go): Bump the go-dependencies group with 7 updates

[dependabot]

Bumps the go-dependencies group with 7 updates:

Package	From	To
github.com/aws/aws-sdk-go-v2	1.43.0	1.43.2
github.com/aws/aws-sdk-go-v2/config	1.32.31	1.32.33
github.com/aws/aws-sdk-go-v2/credentials	1.19.30	1.19.32
github.com/aws/aws-sdk-go-v2/feature/s3/manager	1.22.35	1.22.37
github.com/aws/aws-sdk-go-v2/service/s3	1.106.0	1.106.2
github.com/llm-d/llm-d-async/api	0.7.4	0.9.0
github.com/llm-d/llm-d-async/producer	0.7.4	0.9.0

Updates github.com/aws/aws-sdk-go-v2 from 1.43.0 to 1.43.2

Commits

• 3bf9ef9 Release 2026-07-29
• 3f0db65 Regenerated Clients
• 3212c1c Update endpoints model
• 554a97c Update API model
• 789968e Request compression content length test (#3497)
• 70fc819 generate serde benchmarks (#3492)
• 5cf37fb Release 2026-07-28
• 63ee662 Regenerated Clients
• c4668b7 Update API model
• 8599448 Migrate serde2 kitchensink eventstream tests to smithy-go/eventstream (#3498)
• Additional commits viewable in compare view

Updates github.com/aws/aws-sdk-go-v2/config from 1.32.31 to 1.32.33

Commits

• 3bf9ef9 Release 2026-07-29
• 3f0db65 Regenerated Clients
• 3212c1c Update endpoints model
• 554a97c Update API model
• 789968e Request compression content length test (#3497)
• 70fc819 generate serde benchmarks (#3492)
• 5cf37fb Release 2026-07-28
• 63ee662 Regenerated Clients
• c4668b7 Update API model
• 8599448 Migrate serde2 kitchensink eventstream tests to smithy-go/eventstream (#3498)
• Additional commits viewable in compare view

Updates github.com/aws/aws-sdk-go-v2/credentials from 1.19.30 to 1.19.32

Commits

• 3bf9ef9 Release 2026-07-29
• 3f0db65 Regenerated Clients
• 3212c1c Update endpoints model
• 554a97c Update API model
• 789968e Request compression content length test (#3497)
• 70fc819 generate serde benchmarks (#3492)
• 5cf37fb Release 2026-07-28
• 63ee662 Regenerated Clients
• c4668b7 Update API model
• 8599448 Migrate serde2 kitchensink eventstream tests to smithy-go/eventstream (#3498)
• Additional commits viewable in compare view

Updates github.com/aws/aws-sdk-go-v2/feature/s3/manager from 1.22.35 to 1.22.37

Commits

• 3bf9ef9 Release 2026-07-29
• 3f0db65 Regenerated Clients
• 3212c1c Update endpoints model
• 554a97c Update API model
• 789968e Request compression content length test (#3497)
• 70fc819 generate serde benchmarks (#3492)
• 5cf37fb Release 2026-07-28
• 63ee662 Regenerated Clients
• c4668b7 Update API model
• 8599448 Migrate serde2 kitchensink eventstream tests to smithy-go/eventstream (#3498)
• Additional commits viewable in compare view

Updates github.com/aws/aws-sdk-go-v2/service/s3 from 1.106.0 to 1.106.2

Commits

• 3bf9ef9 Release 2026-07-29
• 3f0db65 Regenerated Clients
• 3212c1c Update endpoints model
• 554a97c Update API model
• 789968e Request compression content length test (#3497)
• 70fc819 generate serde benchmarks (#3492)
• 5cf37fb Release 2026-07-28
• 63ee662 Regenerated Clients
• c4668b7 Update API model
• 8599448 Migrate serde2 kitchensink eventstream tests to smithy-go/eventstream (#3498)
• Additional commits viewable in compare view

Updates github.com/llm-d/llm-d-async/api from 0.7.4 to 0.9.0

Release notes

Sourced from github.com/llm-d/llm-d-async/api's releases.

v0.9.0

What's Changed

• fix(chart): allow modelServerMonitor to scrape across namespaces in llm-d/llm-d-async#359
• fix(chart): apply inference_pool at scrape time when pods lack the label in llm-d/llm-d-async#360
• deps(go): Bump the go-dependencies group with 5 updates in llm-d/llm-d-async#352
• deps(go): Bump k8s.io/client-go from 0.34.9 to 0.34.10 in the kubernetes group in llm-d/llm-d-async#351
• docs: fix stale llm-d paths in the e2e deploy guide in llm-d/llm-d-async#373
• fix(metrics): make pool_name mean one thing across every gate series in llm-d/llm-d-async#372
• docs: explain and surface the prometheus-budget closing point in llm-d/llm-d-async#374
• fix(metrics): count gate_closed where budget gating actually happens in llm-d/llm-d-async#371
• fix(flowcontrol): add a stock-EPP metric source to the budget cascade in llm-d/llm-d-async#370

Full Changelog: llm-d/llm-d-async@v0.8.0...v0.9.0

v0.8.0

---

RELEASE v0.8.0 (2026-07-24)

#300 (2026-07-09) — Breaking: ResultMessage now carries structured result fields so consumers can distinguish HTTP successes, HTTP errors, and non-HTTP failures — added StatusCode (int, non-zero means an HTTP response was received), ErrorCode (string, e.g. DEADLINE_EXCEEDED, GATE_DROPPED, GATE_ERROR, INFERENCE_ERROR, INVALID_REQUEST), and ErrorMessage (string). Wire-format changes to be aware of:

• HTTP-error Payload now contains the raw upstream response body instead of a wrapped error string.
• Gate-drop Payload changed from {"status":"dropped"} to {"error":"Pool gating dropped request"} and is now unified across all transports via NewGateDroppedResult. Detect gate drops via ErrorCode == "GATE_DROPPED" rather than parsing the payload body.
• Gate-drop results now include Metadata (from req.ReqMetadata()), which the previous inline construction omitted.
• InferenceClient.SendRequest now returns (*InferenceResponse, error), where InferenceResponse has StatusCode int and Body []byte fields (previously ([]byte, int, error)).

#306 (2026-07-10) — The default worker concurrency is now 64 (was 8). The processor is I/O-bound, so by Little's Law in-flight concurrency caps throughput; the old default of 8 left real inference pools mostly idle. Tune concurrency to your backend's latency/throughput target (see the Async Processor Operations Guide).

#334 (2026-07-21) — Fixed a gate capacity reservation leak in the Redis sorted-set flow: retried requests never released their per-queue gate reservation, so inFlight ratcheted up on every retry until the queue stopped dispatching entirely (local-max-concurrency) or the redis-quota counter over-admitted on TTL reset. Retries now release the reservation before re-enqueue (and re-reserve on re-dispatch).

#336 (2026-07-21) — Fixed over-admission in concurrency-mode redis-quota: the in-flight counter's TTL was set only on the first acquire and never refreshed, so under sustained load the key expired mid-flight and the counter reset to 0, admitting requests beyond the configured limit. The TTL is now refreshed on every acquire/release, so it only expires after a full window of inactivity (crash-orphan cleanup).

#339 (2026-07-21) — Added two Prometheus gauges (labeled by pool_name) that expose the raw value a metric-based dispatch gate reads and the threshold it compares against, so you can see why a gate is open or closed: async_gate_metric_value (for the saturation gate, 1 - saturation) and async_gate_metric_threshold. Recorded by the prometheus-saturation, prometheus-budget, and prometheus-query gates.

#340 (2026-07-21) — Breaking: the Helm chart is renamed from async-processor to llm-d-async and now publishes to oci://ghcr.io/llm-d/charts/llm-d-async. Rendered resources are named llm-d-async-* with app.kubernetes.io/name=llm-d-async; because a Deployment's selector is immutable, upgrading from an existing async-processor install requires uninstall + reinstall. The old charts/async-processor package remains available (no redirect) but receives no new releases. The default OpenTelemetry trace service name

... (truncated)

Changelog

Sourced from github.com/llm-d/llm-d-async/api's changelog.

Release Notes

User-facing changes for each release, newest first. Release sections are assembled from per-PR fragments in release-notes.d/unreleased/ at each release tag by make release-notes VERSION=vX.Y.Z.

Do not edit release sections by hand. To record a change, add a fragment via the PR template's release-note block — see CONTRIBUTING.md → Release notes.

RELEASE v0.9.0 2026-07-29 2026-07-28 llm-d/llm-d-async#370 Fix the prometheus-budget gate on a stock llm-d install: its primary source queried inference_extension_flow_control_queue_size, which only exists when EPP runs the flow control plugin, so the gate silently served its fallback budget. A source based on inference_pool_per_pod_queue_size (part of EPP's base metric set) is now tried between the flow control and vLLM sources. The resolved PromQL for each cascade tier is logged at startup, and the new llm_d_async_async_gate_metric_source_available gauge distinguishes a fallback budget from a real reading of the same value. 2026-07-28 llm-d/llm-d-async#371 Fix async_gate_decisions_total{reason="gate_closed"}, which could never increment for a budget-based gate: back-pressure is applied pre-dequeue by shrinking the dispatch batch to zero, while the counter was only incremented post-dequeue inside the loop that batch size had just emptied — and a counter series that never increments is absent from /metrics entirely, so the query returned an empty vector rather than 0. The decision is now recorded where it is made: the sorted-set path counts each throttled poll whose batch the budget zeroed while the queue is non-empty, and the Pub/Sub path counts each receive window it skips for the same reason. Both backends also pre-create all four reason series at 0 when a queue starts. The counter's unit is decisions, not messages; the help text and README metric table now say so. 2026-07-29 llm-d/llm-d-async#372 Fix pool_name, which carried three different meanings across the five series an operator needs to diagnose a throttled queue, so none of them joined. async_gate_metric_value and async_gate_metric_threshold labeled it with the gate's pool param — the InferencePool being queried — and carried no queue labels at all; async_dispatch_budget and async_gate_decisions_total read it from the raw queue config, so a queue that omitted worker_pool_id reported pool_name="" while its own async_broker_backlog and async_pool_worker_limit reported "default". pool_name now always names the async worker pool that owns the series, the queue config is normalized before it is stored, and the two gate gauges carry the full queue_id/queue_name/pool_name triple so they join with the rest of the queue's metrics. The InferencePool a gate queries moves to its own inference_pool label, which also makes two worker pools gating on one InferencePool distinguishable. Breaking for dashboards on the two gauges: select {inference_pool="..."} where you previously selected {pool_name="..."}. 2026-07-29 llm-d/llm-d-async#373 Fixed the end-to-end deploy guide (docs/guides/e2e-deploy.md), which referenced llm-d paths and chart flags that no longer exist. Step 5 now installs the llm-d-router-gateway chart instead of the retired GAIE inferencepool chart (httpRoute.* instead of experimentalHttpRoute.*, monitoring via the router recipe's monitoring.values.yaml, pinned to a new ROUTER_CHART_VERSION), the Istio and observability references point at their current locations, and the description of what the install creates now matches the rendered chart. 2026-07-29 llm-d/llm-d-async#374 The prometheus-budget gate now logs its resolved closing point when it is created ("prometheus-budget gate configured" ... closesAtLoadPerReadyPod=95), making a mis-sized max_concurrency visible immediately instead of leaving the gate silently open forever. max_concurrency is documented as a per ready pod capacity in the README and the e2e deploy guide, with guidance on deriving it from the EPP saturation detector's MaxConcurrency or from the observed per-pod peak of vllm:num_requests_running.

RELEASE v0.8.0 2026-07-24 2026-07-09 llm-d/llm-d-async#300 Breaking: ResultMessage now carries structured result fields so consumers can distinguish HTTP successes, HTTP errors, and non-HTTP failures — added StatusCode (int, non-zero means an HTTP response was received), ErrorCode (string, e.g. DEADLINE_EXCEEDED, GATE_DROPPED, GATE_ERROR, INFERENCE_ERROR, INVALID_REQUEST), and ErrorMessage (string). Wire-format changes to be aware of: - HTTP-error Payload now contains the raw upstream response body instead of a wrapped error string. - Gate-drop Payload changed from {"status":"dropped"} to {"error":"Pool gating dropped request"} and is now unified across all transports via NewGateDroppedResult. Detect gate drops via ErrorCode == "GATE_DROPPED" rather than parsing the payload body. - Gate-drop results now include Metadata (from req.ReqMetadata()), which the previous inline construction omitted. - InferenceClient.SendRequest now returns (*InferenceResponse, error), where InferenceResponse has StatusCode int and Body []byte fields (previously ([]byte, int, error)). 2026-07-10 llm-d/llm-d-async#306 The default worker concurrency is now 64 (was 8). The processor is I/O-bound, so by Little's Law in-flight concurrency caps throughput; the old default of 8 left real inference pools mostly idle. Tune concurrency to your backend's latency/throughput target (see the Async Processor Operations Guide). 2026-07-21 llm-d/llm-d-async#334 Fixed a gate capacity reservation leak in the Redis sorted-set flow: retried requests never released their per-queue gate reservation, so inFlight ratcheted up on every retry until the queue stopped dispatching entirely (local-max-concurrency) or the redis-quota counter over-admitted on TTL reset. Retries now release the reservation before re-enqueue (and re-reserve on re-dispatch). 2026-07-21 llm-d/llm-d-async#336 Fixed over-admission in concurrency-mode redis-quota: the in-flight counter's TTL was set only on the first acquire and never refreshed, so under sustained load the key expired mid-flight and the counter reset to 0, admitting requests beyond the configured limit. The TTL is now refreshed on every acquire/release, so it only expires after a full window of inactivity (crash-orphan cleanup). 2026-07-21 llm-d/llm-d-async#339 Added two Prometheus gauges (labeled by pool_name) that expose the raw value a metric-based dispatch gate reads and the threshold it compares against, so you can see why a gate is open or closed: async_gate_metric_value (for the saturation gate, 1 - saturation) and async_gate_metric_threshold. Recorded by the prometheus-saturation, prometheus-budget, and prometheus-query gates. 2026-07-21 llm-d/llm-d-async#340 Breaking: the Helm chart is renamed from async-processor to llm-d-async and now publishes to oci://ghcr.io/llm-d/charts/llm-d-async. Rendered resources are named llm-d-async-* with app.kubernetes.io/name=llm-d-async; because a Deployment's selector is immutable, upgrading from an existing async-processor install requires uninstall + reinstall. The old charts/async-processor package remains available (no redirect) but receives no new releases. The default OpenTelemetry trace service name also changes from async-processor to llm-d-async (OTEL_SERVICE_NAME) — update any trace queries/dashboards accordingly.

Commits

• 877a068 chore(release): prepare v0.9.0 (#375)
• 6839f3c fix(flowcontrol): add a stock-EPP metric source to the budget cascade (#370)
• a69d1dd fix(metrics): count gate_closed where budget gating actually happens (#371)
• 6f93465 docs: explain and surface the prometheus-budget closing point (#374)
• 955feb3 fix(metrics): make pool_name mean one thing across every gate series (#372)
• 0e21aee docs: fix stale llm-d paths in the e2e deploy guide (#373)
• 46ebeb3 deps(go): Bump k8s.io/client-go in the kubernetes group (#351)
• f633234 deps(go): Bump the go-dependencies group with 5 updates (#352)
• 70e5ba6 fix(chart): apply inference_pool at scrape time when pods lack the label (#360)
• 2dd025b fix(chart): allow modelServerMonitor to scrape across namespaces (#359)
• Additional commits viewable in compare view

Updates github.com/llm-d/llm-d-async/producer from 0.7.4 to 0.9.0

Release notes

Sourced from github.com/llm-d/llm-d-async/producer's releases.

v0.9.0

What's Changed

• fix(chart): allow modelServerMonitor to scrape across namespaces in llm-d/llm-d-async#359
• fix(chart): apply inference_pool at scrape time when pods lack the label in llm-d/llm-d-async#360
• deps(go): Bump the go-dependencies group with 5 updates in llm-d/llm-d-async#352
• deps(go): Bump k8s.io/client-go from 0.34.9 to 0.34.10 in the kubernetes group in llm-d/llm-d-async#351
• docs: fix stale llm-d paths in the e2e deploy guide in llm-d/llm-d-async#373
• fix(metrics): make pool_name mean one thing across every gate series in llm-d/llm-d-async#372
• docs: explain and surface the prometheus-budget closing point in llm-d/llm-d-async#374
• fix(metrics): count gate_closed where budget gating actually happens in llm-d/llm-d-async#371
• fix(flowcontrol): add a stock-EPP metric source to the budget cascade in llm-d/llm-d-async#370

Full Changelog: llm-d/llm-d-async@v0.8.0...v0.9.0

v0.8.0

---

RELEASE v0.8.0 (2026-07-24)

#300 (2026-07-09) — Breaking: ResultMessage now carries structured result fields so consumers can distinguish HTTP successes, HTTP errors, and non-HTTP failures — added StatusCode (int, non-zero means an HTTP response was received), ErrorCode (string, e.g. DEADLINE_EXCEEDED, GATE_DROPPED, GATE_ERROR, INFERENCE_ERROR, INVALID_REQUEST), and ErrorMessage (string). Wire-format changes to be aware of:

• HTTP-error Payload now contains the raw upstream response body instead of a wrapped error string.
• Gate-drop Payload changed from {"status":"dropped"} to {"error":"Pool gating dropped request"} and is now unified across all transports via NewGateDroppedResult. Detect gate drops via ErrorCode == "GATE_DROPPED" rather than parsing the payload body.
• Gate-drop results now include Metadata (from req.ReqMetadata()), which the previous inline construction omitted.
• InferenceClient.SendRequest now returns (*InferenceResponse, error), where InferenceResponse has StatusCode int and Body []byte fields (previously ([]byte, int, error)).

#306 (2026-07-10) — The default worker concurrency is now 64 (was 8). The processor is I/O-bound, so by Little's Law in-flight concurrency caps throughput; the old default of 8 left real inference pools mostly idle. Tune concurrency to your backend's latency/throughput target (see the Async Processor Operations Guide).

#334 (2026-07-21) — Fixed a gate capacity reservation leak in the Redis sorted-set flow: retried requests never released their per-queue gate reservation, so inFlight ratcheted up on every retry until the queue stopped dispatching entirely (local-max-concurrency) or the redis-quota counter over-admitted on TTL reset. Retries now release the reservation before re-enqueue (and re-reserve on re-dispatch).

#336 (2026-07-21) — Fixed over-admission in concurrency-mode redis-quota: the in-flight counter's TTL was set only on the first acquire and never refreshed, so under sustained load the key expired mid-flight and the counter reset to 0, admitting requests beyond the configured limit. The TTL is now refreshed on every acquire/release, so it only expires after a full window of inactivity (crash-orphan cleanup).

#339 (2026-07-21) — Added two Prometheus gauges (labeled by pool_name) that expose the raw value a metric-based dispatch gate reads and the threshold it compares against, so you can see why a gate is open or closed: async_gate_metric_value (for the saturation gate, 1 - saturation) and async_gate_metric_threshold. Recorded by the prometheus-saturation, prometheus-budget, and prometheus-query gates.

#340 (2026-07-21) — Breaking: the Helm chart is renamed from async-processor to llm-d-async and now publishes to oci://ghcr.io/llm-d/charts/llm-d-async. Rendered resources are named llm-d-async-* with app.kubernetes.io/name=llm-d-async; because a Deployment's selector is immutable, upgrading from an existing async-processor install requires uninstall + reinstall. The old charts/async-processor package remains available (no redirect) but receives no new releases. The default OpenTelemetry trace service name

... (truncated)

Changelog

Sourced from github.com/llm-d/llm-d-async/producer's changelog.

Release Notes

User-facing changes for each release, newest first. Release sections are assembled from per-PR fragments in release-notes.d/unreleased/ at each release tag by make release-notes VERSION=vX.Y.Z.

Do not edit release sections by hand. To record a change, add a fragment via the PR template's release-note block — see CONTRIBUTING.md → Release notes.

RELEASE v0.9.0 2026-07-29 2026-07-28 llm-d/llm-d-async#370 Fix the prometheus-budget gate on a stock llm-d install: its primary source queried inference_extension_flow_control_queue_size, which only exists when EPP runs the flow control plugin, so the gate silently served its fallback budget. A source based on inference_pool_per_pod_queue_size (part of EPP's base metric set) is now tried between the flow control and vLLM sources. The resolved PromQL for each cascade tier is logged at startup, and the new llm_d_async_async_gate_metric_source_available gauge distinguishes a fallback budget from a real reading of the same value. 2026-07-28 llm-d/llm-d-async#371 Fix async_gate_decisions_total{reason="gate_closed"}, which could never increment for a budget-based gate: back-pressure is applied pre-dequeue by shrinking the dispatch batch to zero, while the counter was only incremented post-dequeue inside the loop that batch size had just emptied — and a counter series that never increments is absent from /metrics entirely, so the query returned an empty vector rather than 0. The decision is now recorded where it is made: the sorted-set path counts each throttled poll whose batch the budget zeroed while the queue is non-empty, and the Pub/Sub path counts each receive window it skips for the same reason. Both backends also pre-create all four reason series at 0 when a queue starts. The counter's unit is decisions, not messages; the help text and README metric table now say so. 2026-07-29 llm-d/llm-d-async#372 Fix pool_name, which carried three different meanings across the five series an operator needs to diagnose a throttled queue, so none of them joined. async_gate_metric_value and async_gate_metric_threshold labeled it with the gate's pool param — the InferencePool being queried — and carried no queue labels at all; async_dispatch_budget and async_gate_decisions_total read it from the raw queue config, so a queue that omitted worker_pool_id reported pool_name="" while its own async_broker_backlog and async_pool_worker_limit reported "default". pool_name now always names the async worker pool that owns the series, the queue config is normalized before it is stored, and the two gate gauges carry the full queue_id/queue_name/pool_name triple so they join with the rest of the queue's metrics. The InferencePool a gate queries moves to its own inference_pool label, which also makes two worker pools gating on one InferencePool distinguishable. Breaking for dashboards on the two gauges: select {inference_pool="..."} where you previously selected {pool_name="..."}. 2026-07-29 llm-d/llm-d-async#373 Fixed the end-to-end deploy guide (docs/guides/e2e-deploy.md), which referenced llm-d paths and chart flags that no longer exist. Step 5 now installs the llm-d-router-gateway chart instead of the retired GAIE inferencepool chart (httpRoute.* instead of experimentalHttpRoute.*, monitoring via the router recipe's monitoring.values.yaml, pinned to a new ROUTER_CHART_VERSION), the Istio and observability references point at their current locations, and the description of what the install creates now matches the rendered chart. 2026-07-29 llm-d/llm-d-async#374 The prometheus-budget gate now logs its resolved closing point when it is created ("prometheus-budget gate configured" ... closesAtLoadPerReadyPod=95), making a mis-sized max_concurrency visible immediately instead of leaving the gate silently open forever. max_concurrency is documented as a per ready pod capacity in the README and the e2e deploy guide, with guidance on deriving it from the EPP saturation detector's MaxConcurrency or from the observed per-pod peak of vllm:num_requests_running.

RELEASE v0.8.0 2026-07-24 2026-07-09 llm-d/llm-d-async#300 Breaking: ResultMessage now carries structured result fields so consumers can distinguish HTTP successes, HTTP errors, and non-HTTP failures — added StatusCode (int, non-zero means an HTTP response was received), ErrorCode (string, e.g. DEADLINE_EXCEEDED, GATE_DROPPED, GATE_ERROR, INFERENCE_ERROR, INVALID_REQUEST), and ErrorMessage (string). Wire-format changes to be aware of: - HTTP-error Payload now contains the raw upstream response body instead of a wrapped error string. - Gate-drop Payload changed from {"status":"dropped"} to {"error":"Pool gating dropped request"} and is now unified across all transports via NewGateDroppedResult. Detect gate drops via ErrorCode == "GATE_DROPPED" rather than parsing the payload body. - Gate-drop results now include Metadata (from req.ReqMetadata()), which the previous inline construction omitted. - InferenceClient.SendRequest now returns (*InferenceResponse, error), where InferenceResponse has StatusCode int and Body []byte fields (previously ([]byte, int, error)). 2026-07-10 llm-d/llm-d-async#306 The default worker concurrency is now 64 (was 8). The processor is I/O-bound, so by Little's Law in-flight concurrency caps throughput; the old default of 8 left real inference pools mostly idle. Tune concurrency to your backend's latency/throughput target (see the Async Processor Operations Guide). 2026-07-21 llm-d/llm-d-async#334 Fixed a gate capacity reservation leak in the Redis sorted-set flow: retried requests never released their per-queue gate reservation, so inFlight ratcheted up on every retry until the queue stopped dispatching entirely (local-max-concurrency) or the redis-quota counter over-admitted on TTL reset. Retries now release the reservation before re-enqueue (and re-reserve on re-dispatch). 2026-07-21 llm-d/llm-d-async#336 Fixed over-admission in concurrency-mode redis-quota: the in-flight counter's TTL was set only on the first acquire and never refreshed, so under sustained load the key expired mid-flight and the counter reset to 0, admitting requests beyond the configured limit. The TTL is now refreshed on every acquire/release, so it only expires after a full window of inactivity (crash-orphan cleanup). 2026-07-21 llm-d/llm-d-async#339 Added two Prometheus gauges (labeled by pool_name) that expose the raw value a metric-based dispatch gate reads and the threshold it compares against, so you can see why a gate is open or closed: async_gate_metric_value (for the saturation gate, 1 - saturation) and async_gate_metric_threshold. Recorded by the prometheus-saturation, prometheus-budget, and prometheus-query gates. 2026-07-21 llm-d/llm-d-async#340 Breaking: the Helm chart is renamed from async-processor to llm-d-async and now publishes to oci://ghcr.io/llm-d/charts/llm-d-async. Rendered resources are named llm-d-async-* with app.kubernetes.io/name=llm-d-async; because a Deployment's selector is immutable, upgrading from an existing async-processor install requires uninstall + reinstall. The old charts/async-processor package remains available (no redirect) but receives no new releases. The default OpenTelemetry trace service name also changes from async-processor to llm-d-async (OTEL_SERVICE_NAME) — update any trace queries/dashboards accordingly.

Commits

• 877a068 chore(release): prepare v0.9.0 (#375)
• 6839f3c fix(flowcontrol): add a stock-EPP metric source to the budget cascade (#370)
• a69d1dd fix(metrics): count gate_closed where budget gating actually happens (#371)
• 6f93465 docs: explain and surface the prometheus-budget closing point (#374)
• 955feb3 fix(metrics): make pool_name mean one thing across every gate series (#372)
• 0e21aee docs: fix stale llm-d paths in the e2e deploy guide (#373)
• 46ebeb3 deps(go): Bump k8s.io/client-go in the kubernetes group (#351)
• f633234 deps(go): Bump the go-dependencies group with 5 updates (#352)
• 70e5ba6 fix(chart): apply inference_pool at scrape time when pods lack the label (#360)
• 2dd025b fix(chart): allow modelServerMonitor to scrape across namespaces (#359)
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore <dependency name> major version will close this group update PR and stop Dependabot creating any more for the specific dependency's major version (unless you unignore this specific dependency's major version or upgrade to it yourself)
• @dependabot ignore <dependency name> minor version will close this group update PR and stop Dependabot creating any more for the specific dependency's minor version (unless you unignore this specific dependency's minor version or upgrade to it yourself)
• @dependabot ignore <dependency name> will close this group update PR and stop Dependabot creating any more for the specific dependency (unless you unignore this specific dependency or upgrade to it yourself)
• @dependabot unignore <dependency name> will remove all of the ignore conditions of the specified dependency
• @dependabot unignore <dependency name> <ignore condition> will remove the ignore condition of the specified dependency and ignore conditions

[github-actions]

Unsigned commits detected! Please sign your commits.

For instructions on how to set up GPG/SSH signing and verify your commits, please see GitHub Documentation.