diff --git a/deploy/hive-oke/README.md b/deploy/hive-oke/README.md index feb8954..0583920 100644 --- a/deploy/hive-oke/README.md +++ b/deploy/hive-oke/README.md @@ -21,7 +21,7 @@ Quorum (RFC #1) uses one shared Vexa control plane per cluster. On hive-oke it r ## Versions (all pinned by tag and digest) -Everything Vexa comes from one commit of hivecommons/vexa `main`: 3addb0a940acc42e899b2748711c93dab2467874 (chart 0.13.4). +Everything Vexa comes from one commit of hivecommons/vexa `main`: 93bc10a5d2972197642a960e885b356a5063d50a (chart 0.13.4). It is upstream Vexa `main` plus our patches (hivecommons/vexa#2 Teams chat-write, #3 speak route, #7 Teams per-track capture, #8 Google Meet end of turn, #9 cached and prewarmed short TTS lines). Upstream v0.12.28 has no meeting chat-post route and its `/ws` sends no chat events; both @@ -29,15 +29,15 @@ are on upstream `main`, and upstream publishes control-plane images only for rel | Component | Image | | ---------------------------------------- | -------------------------------------------------------------------------------- | -| Chart | `deploy/helm/charts/vexa` at hivecommons/vexa 3addb0a (chart 0.13.4) | -| gateway | `ghcr.io/hivecommons/vexa-gateway:main-3addb0a-quorum` | -| admin-api | `ghcr.io/hivecommons/vexa-admin-api:main-3addb0a-quorum` | -| meeting-api | `ghcr.io/hivecommons/vexa-meeting-api:main-3addb0a-quorum` | -| runtime | `ghcr.io/hivecommons/vexa-runtime:main-3addb0a-quorum` | -| meeting bot | `ghcr.io/hivecommons/vexa-bot:main-3addb0a-quorum` | -| CPU transcription (hotwords, warm-up) | `ghcr.io/hivecommons/vexa-transcription-cpu:main-3addb0a-quorum` | -| agent-api, agent-worker (pinned, unused) | `ghcr.io/hivecommons/vexa-agent-api`, `vexa-agent-worker` `:main-3addb0a-quorum` | -| mcp (pinned, disabled) | `ghcr.io/hivecommons/vexa-mcp:main-3addb0a-quorum` | +| Chart | `deploy/helm/charts/vexa` at hivecommons/vexa 93bc10a (chart 0.13.4) | +| gateway | `ghcr.io/hivecommons/vexa-gateway:main-93bc10a-quorum` | +| admin-api | `ghcr.io/hivecommons/vexa-admin-api:main-93bc10a-quorum` | +| meeting-api | `ghcr.io/hivecommons/vexa-meeting-api:main-93bc10a-quorum` | +| runtime | `ghcr.io/hivecommons/vexa-runtime:main-93bc10a-quorum` | +| meeting bot | `ghcr.io/hivecommons/vexa-bot:main-93bc10a-quorum` | +| CPU transcription (hotwords, warm-up) | `ghcr.io/hivecommons/vexa-transcription-cpu:main-93bc10a-quorum` | +| agent-api, agent-worker (pinned, unused) | `ghcr.io/hivecommons/vexa-agent-api`, `vexa-agent-worker` `:main-93bc10a-quorum` | +| mcp (pinned, disabled) | `ghcr.io/hivecommons/vexa-mcp:main-93bc10a-quorum` | | postgres | `docker.io/library/postgres:17.11-alpine` | | redis | `docker.io/valkey/valkey:8.1.11-alpine` | | S3 | `docker.io/versity/versitygw:v1.10.0` | @@ -233,7 +233,7 @@ the upstream Kubernetes docs, it is minted at install time and kept in `vexa-adm ```bash git clone https://github.com/hivecommons/vexa.git /tmp/hivecommons-vexa -git -C /tmp/hivecommons-vexa checkout 3addb0a940acc42e899b2748711c93dab2467874 # quorum.vexaCommit in values.yaml +git -C /tmp/hivecommons-vexa checkout 93bc10a5d2972197642a960e885b356a5063d50a # quorum.vexaCommit in values.yaml KUBE_CONTEXT=hive-oke VEXA_CHART=/tmp/hivecommons-vexa/deploy/helm/charts/vexa ./install.sh KUBE_CONTEXT=hive-oke ./smoke.sh ``` diff --git a/deploy/hive-oke/transcription.yaml b/deploy/hive-oke/transcription.yaml index b57bf73..ac95dff 100644 --- a/deploy/hive-oke/transcription.yaml +++ b/deploy/hive-oke/transcription.yaml @@ -40,8 +40,8 @@ spec: automountServiceAccountToken: false containers: - name: transcription - # main-3addb0a-quorum (linux/amd64, linux/arm64), hivecommons/vexa 3addb0a (hotwords, warm-up) - image: ghcr.io/hivecommons/vexa-transcription-cpu:main-3addb0a-quorum@sha256:5dd732b156a00813e7b02e4d30f2e851647e689ef0f9eb29375d4c511079b097 + # main-93bc10a-quorum (linux/amd64, linux/arm64), hivecommons/vexa 93bc10a (hotwords, warm-up) + image: ghcr.io/hivecommons/vexa-transcription-cpu:main-93bc10a-quorum@sha256:37158bdcdffe378060994b07135759b782bb6aff6f38e0e36ab53c1321309d50 env: - name: WORKER_ID value: "1" @@ -118,7 +118,7 @@ spec: # Fetch faster-whisper `base.en` at a pinned commit of Systran/faster-whisper-base.en into the # model cache, and point the cache's refs/main at it so the offline server resolves it. - name: fetch-model - image: ghcr.io/hivecommons/vexa-transcription-cpu:main-3addb0a-quorum@sha256:5dd732b156a00813e7b02e4d30f2e851647e689ef0f9eb29375d4c511079b097 + image: ghcr.io/hivecommons/vexa-transcription-cpu:main-93bc10a-quorum@sha256:37158bdcdffe378060994b07135759b782bb6aff6f38e0e36ab53c1321309d50 env: - name: MODEL_REVISION value: 3d3d5dee26484f91867d81cb899cfcf72b96be6c @@ -148,7 +148,7 @@ spec: containers: - name: transcription # Same image and digest as the batch tier above. - image: ghcr.io/hivecommons/vexa-transcription-cpu:main-3addb0a-quorum@sha256:5dd732b156a00813e7b02e4d30f2e851647e689ef0f9eb29375d4c511079b097 + image: ghcr.io/hivecommons/vexa-transcription-cpu:main-93bc10a-quorum@sha256:37158bdcdffe378060994b07135759b782bb6aff6f38e0e36ab53c1321309d50 env: - name: WORKER_ID value: "rt" diff --git a/deploy/hive-oke/values.yaml b/deploy/hive-oke/values.yaml index 2e69bc2..536a683 100644 --- a/deploy/hive-oke/values.yaml +++ b/deploy/hive-oke/values.yaml @@ -1,6 +1,6 @@ # Vexa Helm values for the shared Quorum meeting service on hive-oke (namespace `meetings`). # -# Chart: deploy/helm/charts/vexa from hivecommons/vexa main at commit 3addb0a940acc42e899b2748711c93dab2467874 (chart 0.13.4, +# Chart: deploy/helm/charts/vexa from hivecommons/vexa main at commit 93bc10a5d2972197642a960e885b356a5063d50a (chart 0.13.4, # appVersion 0.13.4). Install with ./install.sh, which also creates the Secrets and applies the # namespace policy rendered from the `quorum` block below. # @@ -18,7 +18,7 @@ # ---- Quorum deployment knobs (read by install.sh, ignored by the chart) ---- quorum: # The hivecommons/vexa commit the chart is taken from and every Vexa image is built from. - vexaCommit: 3addb0a940acc42e899b2748711c93dab2467874 + vexaCommit: 93bc10a5d2972197642a960e885b356a5063d50a # Most meeting-bot Pods that may run at once in `meetings`. Enforced twice: # 1. ResourceQuota `meetings-quota`: pods = controlPlanePods + maxConcurrentBots # (bot Pods are bare Pods the runtime creates, so a quota cannot select only them); @@ -72,13 +72,13 @@ mcp: enabled: false image: repository: ghcr.io/hivecommons/vexa-mcp - tag: main-3addb0a-quorum@sha256:d4f802bce02ff8e22022b4f708651d6ba2459772f231b54d09fd73f037d3a84e + tag: main-93bc10a-quorum@sha256:dcd8ecc676c11e556abfbd4782e49dc5418f871127d2a05c5d8e8d553bb64611 gateway: replicaCount: 2 image: repository: ghcr.io/hivecommons/vexa-gateway - tag: main-3addb0a-quorum@sha256:ac40a5d374a8fef37dd4df7a31dc97b83e56555e5b636bf42e58a6d3f4a6691e + tag: main-93bc10a-quorum@sha256:76803e7958da25f9b989b554773d65bc8390c3d291ee80c10a0756545e2e632d service: type: ClusterIP @@ -86,13 +86,13 @@ adminApi: replicaCount: 2 image: repository: ghcr.io/hivecommons/vexa-admin-api - tag: main-3addb0a-quorum@sha256:5ecf337d6d63c415bc2a0ce6a78e02df8c71e3dc8ecc28cc9486bef3c22a0e63 + tag: main-93bc10a-quorum@sha256:0db02dc2551139e6b74561839f2e621eb73570e49cb702e94c97d4dea1ed323b meetingApi: replicaCount: 2 image: repository: ghcr.io/hivecommons/vexa-meeting-api - tag: main-3addb0a-quorum@sha256:6f7e881be4f643bdd3b860c50635270e2c3d597f7ad80b476532bee832b3e569 + tag: main-93bc10a-quorum@sha256:bd6a7aeaa034c1dd760aa84c79588c783845bd8c73302385384b74795688edf8 # Participants see this name; it says the bot takes notes (RFC: behaviour in the meeting). defaultBotName: "Hive notes (AI notetaker)" transcriptionServiceUrl: http://vexa-transcription:8000 @@ -117,16 +117,16 @@ runtime: replicaCount: 1 image: repository: ghcr.io/hivecommons/vexa-runtime - tag: main-3addb0a-quorum@sha256:d5a51457840a0e349415e266aa21ad706a75daed9c640c645454b2a620ee3529 + tag: main-93bc10a-quorum@sha256:1c815b101a91dfd6b269030bc1a6223d708043efe9eadb533eb4f116cd51b020 # The meeting bot, from the same commit as everything else (hivecommons/vexa#15: lines are # synthesized and played sentence by sentence from a per-sentence cache, the microphone is # opened once at join and carries only the speech clip, the Meet final follows the pause by # one transcription round trip, Teams utterances cross whole and Teams chat text is typed in # and checked before send). install.sh also pre-pulls this image on every bot node. - browserImage: ghcr.io/hivecommons/vexa-bot:main-3addb0a-quorum@sha256:5ae0a60e4123d7e236b9896f882005201273dc9eb474582b955e9d685d28bda7 + browserImage: ghcr.io/hivecommons/vexa-bot:main-93bc10a-quorum@sha256:2170dbde3c706cea6214a925c2b89c2341b0b9fd1eebc89ba95f829fb44dd511 # Agent workers are not used by Quorum (agentApi is off); pinned anyway so nothing is mutable. - agentImage: ghcr.io/hivecommons/vexa-agent-api:main-3addb0a-quorum@sha256:f3e38905b1ed0c2d53e5cc8c205b1d27349b30e5c74df842d01cf6ea2ec18de3 - agentWorkerImage: ghcr.io/hivecommons/vexa-agent-worker:main-3addb0a-quorum@sha256:4bb8d561bd1e386c2ef3d1539437e133e74eb3342056da3366fe03570e74ddf0 + agentImage: ghcr.io/hivecommons/vexa-agent-api:main-93bc10a-quorum@sha256:86db5f578b637f248f200ca9a7f88c794179b0c63a9389fbe944c5a689c6aac9 + agentWorkerImage: ghcr.io/hivecommons/vexa-agent-worker:main-93bc10a-quorum@sha256:c40bd70cfb7721036398e0f14e7e529de5b2a16fd52594aff9248903b7a2470f # Each spawned bot Pod: Guaranteed QoS, request == limit. Sizing per upstream production # guidance (deploy docs: ~1.0-1.3Gi steady, Teams spikes past 2Gi; do not go below 2560Mi). workloadResources: