Tags: Max-NV/nvcf
Tags
feat(llm-api-gateway): add function ID metric labels (NVIDIA#237) * feat(llm-api-gateway): add function ID metric labels Expose routing-key-derived function_id labels on request, upstream, token, and latency metrics so operators can aggregate per-function behavior. Use none for unrouted endpoints and keep infrastructure metrics function-independent. Signed-off-by: jcameron <jcameron@nvidia.com> * docs(llm-api-gateway): document function metric labels Document the request-facing metric names, function_id semantics, and example PromQL queries so operators can monitor per-function traffic and latency. Signed-off-by: jcameron <jcameron@nvidia.com> * fix(llm-api-gateway): address metrics review feedback Add function attribution to request and stream spans, handle missing upstream request context safely, and declare the direct Bazel test dependency. Signed-off-by: jcameron <jcameron@nvidia.com> --------- Signed-off-by: jcameron <jcameron@nvidia.com>
fix(function-autoscaler): avoid startup health panic (NVIDIA#220) Signed-off-by: Bora Oztekin <boztekin@nvidia.com>
feat(openbao-migrations): support optional JWT issuer override (NVIDI… …A#236) Add an optional JWT_ISSUER_OVERRIDE input that, when set, overrides the derived in-cluster issuer ("http://<service>.<namespace>.svc.cluster.local") stamped into the generated JWT secret roles. Leaving it empty preserves the existing per-service default, so current deployments are unaffected. Also build the JWT secret role JSON with jq (--arg/--argjson) instead of a heredoc, so every string value is safely escaped (quotes, backslashes, and control characters) rather than interpolated directly.
fix(migrations): skip non-directory entries under keyspaces/ (NVIDIA#192 ) migrations/cassandra/keyspaces/ holds README.md alongside the four keyspace directories. execute_sqls.sh iterated over every entry, so README.md was treated as a keyspace and passed to migrate as x-migrations-table=README.md. The golang-migrate Cassandra driver reads the dot in that value as keyspace.table, so it tried to open a keyspace named README, which Cassandra lowercases: error: failed to open database: Keyspace 'readme' doesn't exist Migration failed for /tmp/keyspaces/README.md execute_sqls.sh then exits 1, which fails the Helm hook job and leaves the cassandra release in a failed state. Guard both loops with [ -d ] so only directories are treated as keyspaces. This also covers any other non-directory file added to keyspaces/ later. Fixes NVIDIA#235 Signed-off-by: Suresh Krishnan <sureshkrishnan.ai@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
feat(llm-api-gateway): add function ID metric labels (NVIDIA#237) * feat(llm-api-gateway): add function ID metric labels Expose routing-key-derived function_id labels on request, upstream, token, and latency metrics so operators can aggregate per-function behavior. Use none for unrouted endpoints and keep infrastructure metrics function-independent. Signed-off-by: jcameron <jcameron@nvidia.com> * docs(llm-api-gateway): document function metric labels Document the request-facing metric names, function_id semantics, and example PromQL queries so operators can monitor per-function traffic and latency. Signed-off-by: jcameron <jcameron@nvidia.com> * fix(llm-api-gateway): address metrics review feedback Add function attribution to request and stream spans, handle missing upstream request context safely, and declare the direct Bazel test dependency. Signed-off-by: jcameron <jcameron@nvidia.com> --------- Signed-off-by: jcameron <jcameron@nvidia.com>
docs(agents): remove dead skill and file references (NVIDIA#201) AGENTS.md, ai-tooling/README.md, and BAZEL.md listed several skills that never existed on disk, and pointed at deploy/helm/AGENTS.md, migrations/AGENTS.md, and dependencies-java.md, none of which were ever created. Drop the dead entries so the guidance matches the repo. NO-REF Signed-off-by: mesutoezdil <mesudozdil@gmail.com>
fix(function-autoscaler): avoid startup health panic (NVIDIA#220) Signed-off-by: Bora Oztekin <boztekin@nvidia.com>
feat(openbao-migrations): support optional JWT issuer override (NVIDI… …A#236) Add an optional JWT_ISSUER_OVERRIDE input that, when set, overrides the derived in-cluster issuer ("http://<service>.<namespace>.svc.cluster.local") stamped into the generated JWT secret roles. Leaving it empty preserves the existing per-service default, so current deployments are unaffected. Also build the JWT secret role JSON with jq (--arg/--argjson) instead of a heredoc, so every string value is safely escaped (quotes, backslashes, and control characters) rather than interpolated directly.
fix(migrations): skip non-directory entries under keyspaces/ (NVIDIA#192 ) migrations/cassandra/keyspaces/ holds README.md alongside the four keyspace directories. execute_sqls.sh iterated over every entry, so README.md was treated as a keyspace and passed to migrate as x-migrations-table=README.md. The golang-migrate Cassandra driver reads the dot in that value as keyspace.table, so it tried to open a keyspace named README, which Cassandra lowercases: error: failed to open database: Keyspace 'readme' doesn't exist Migration failed for /tmp/keyspaces/README.md execute_sqls.sh then exits 1, which fails the Helm hook job and leaves the cassandra release in a failed state. Guard both loops with [ -d ] so only directories are treated as keyspaces. This also covers any other non-directory file added to keyspaces/ later. Fixes NVIDIA#235 Signed-off-by: Suresh Krishnan <sureshkrishnan.ai@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
feat(openbao-migrations): support optional JWT issuer override (NVIDI… …A#236) Add an optional JWT_ISSUER_OVERRIDE input that, when set, overrides the derived in-cluster issuer ("http://<service>.<namespace>.svc.cluster.local") stamped into the generated JWT secret roles. Leaving it empty preserves the existing per-service default, so current deployments are unaffected. Also build the JWT secret role JSON with jq (--arg/--argjson) instead of a heredoc, so every string value is safely escaped (quotes, backslashes, and control characters) rather than interpolated directly.
PreviousNext