4 Commits

Include renames

Author	SHA1	Message	Date
krishicks	0b351c4a9b	fix(helm)!: reduce gateway Secret privileges (#3616) ... * fix(driver-kubernetes-secrets)!: store provider credentials in one namespace The Kubernetes Secrets credential driver now stores every credential in its configured namespace in all workspace modes and rejects handles that reference any other namespace before contacting the Kubernetes API. The gateway reaches credential Secrets through the Role in that namespace; this allows removing the Secret rules from the ClusterRole. - Remove the workspace_mode, gateway_id, and allow_reference_namespace driver settings and stop rendering them from Helm. Configurations that set them fail at startup. Existing credential state is not migrated. - Add server.credentialDrivers.kubernetesSecrets.createNamespace to provision a dedicated credential namespace. The namespace is kept on uninstall, adopted by a reinstall of the same release, and left untouched when something else owns it. - Update the gateway config reference, Kubernetes setup docs, 0.1.0 upgrade guide, compute-runtime architecture, and cluster debugging skill. Signed-off-by: Kris Hicks <khicks@nvidia.com> * fix(helm): reduce gateway Secret permissions Remove the gateway's Secret list permission in every workspace mode and grant source Secret reads through a Role in the sandbox namespace. Bootstrap Secret cleanup deletes Secrets by exact name instead of listing them. - Grant get on the copied client TLS and image-pull Secrets through a Role in the sandbox namespace. The ClusterRole keeps get and patch on those names for the ownership check and server-side apply into workspace namespaces. - Delete sandbox and supervisor bootstrap Secrets by exact name, derived from the runtime generation recorded on the Sandbox and, on restart, the target generation, tolerating 404. The generation annotation is cleared only after cleanup succeeds, and each bootstrap Secret has a Pod owner reference, so garbage collection removes any generation the driver does not name. - Drop Secret list from the ClusterRole and the shared-mode sandbox Role. - Extend the managed e2e RBAC checks to Secret list. - Update the Kubernetes setup and sandbox runtime docs, compute-runtime architecture, and cluster debugging skill. Signed-off-by: Kris Hicks <khicks@nvidia.com> * fix(driver-kubernetes): stage workspace Secrets per runtime generation Every Secret the Kubernetes driver writes into a workspace namespace is now scoped to one sandbox runtime generation, immutable, and created with create only, so the gateway never reads, patches, or adopts an existing Secret there. This removes the gateway's cluster-wide get and patch on the copied client TLS and image-pull Secret names. - In managed mode, create an immutable copy of each configured image-pull Secret per generation, named os-pull-<id>-<generation>-<n> and owned by the generation's workload and supervisor Pods. Pods and the restarted Sandbox template reference those names, and generation cleanup deletes them by name. A Secret already holding a generation name fails the create. - Outside shared mode, stage the gateway client TLS material into the supervisor bootstrap Secret instead of copying the client TLS Secret into the workspace namespace. - Remove the fixed-name TLS and image-pull copies, the target ownership read, and the ClusterRole get and patch rule on the copied names. Source reads stay in the sandbox-namespace Role. - Update the managed e2e to expect generation image-pull Secrets and client TLS material in the supervisor bootstrap Secret, and to check that the gateway cannot read the copied names in workspace namespaces. - Update the gateway config and compute driver references, Kubernetes setup and sandbox runtime docs, compute-runtime architecture, driver README, and cluster debugging skill. Signed-off-by: Kris Hicks <khicks@nvidia.com> * fix(helm)!: grant operator-mode Secret permissions through the workspace chart The operator-mode gateway ClusterRole grants no Secret permissions. The openshell-workspace chart Role, installed in each operator-managed namespace, grants the gateway create and delete on Secrets for sandbox runtime generations. Operator-managed namespaces require the workspace chart. - Fail the chart tests on any ClusterRole rule that includes Secrets in operator and shared modes. - Install the workspace chart when the operator e2e provisions a namespace, and assert that the gateway has no Secret permissions in a namespace without it. - Update the Kubernetes setup docs, 0.1.0 upgrade guide, compute-runtime architecture, and cluster debugging skill. Signed-off-by: Kris Hicks <khicks@nvidia.com> --------- Signed-off-by: Kris Hicks <khicks@nvidia.com>	2026-09-23 21:35:22 +00:00
4 people	0a770d9173	feat(kubernetes): support HA gateway rebalancing (#1868) ... * feat(kubernetes): support HA gateway rebalancing Signed-off-by: Drew Newberry <anewberry@nvidia.com> * perf(server): cache peer connections, tokens, and owner lookups Every forwarded relay rebuilt its setup from scratch: an owner lookup, a blocking read of the peer token, a TLS connect to the owning replica, and a TokenReview plus Pod GET on the receiving side. Sandbox service routing does this per HTTP request, so the apiserver calls scaled with traffic. Cache all of it on ServerState: - peer channels pooled per endpoint, so relays multiplex over one connection instead of redialing - peer tokens keyed by SHA-256, expiring at min(ttl, token exp) so a hit cannot accept an expired token - owner records for 3s against a 45s ownership TTL, still freshness checked before use Entries are evicted when a relay fails. Also raise HTTP/2 max_concurrent_streams to 1024, since pooling funnels every relay between two replicas onto one connection and hyper's default of 200 sits below the 256 pending-relay budget. Signed-off-by: divesh <dgude@nvidia.com> * perf(server): pool upstream connections for sandbox services Each HTTP request to a sandbox service opened its own supervisor relay, paying a new TCP connection and HTTP/1 handshake every time. Worse, it counted against the 32 in-flight relay cap, so a service handling more than 32 concurrent requests failed outright. Pool idle upstreams per endpoint and port, up to 8 each for 15s. Reuse is safe because the pool only returns a connection hyper reports as ready, and HTTP/1 cannot start a request until the previous body has drained. Upgrades are never pooled since they take the connection over, and a failed send evicts that endpoint. Pruning is bounded per key, with the full sweep limited to once per 30s. Signed-off-by: divesh <dgude@nvidia.com> * fix(server): address HA gateway review findings (#3449) - Let a gateway own supervisor sessions without a peer endpoint. Requiring one whenever the store is PostgreSQL broke every single-instance PostgreSQL deployment, because no sandbox supervisor could connect. A cross-replica request to an owner that advertises no endpoint now fails immediately naming the cause, instead of retrying until the wait timeout. - Close a supervisor session on heartbeat only when another replica owns it, or after renewals fail for the ownership TTL. A database error no longer drops every session heartbeating during an outage. - Clamp owner record ages at zero so a skewed or corrupt stored timestamp cannot produce a negative age. - Bound the cross-object advisory lock with a lock timeout, so a stuck holder fails instead of blocking every mutation in the fleet. - Refuse to start when a peer endpoint is configured on a multi-replica backend but peer authentication is unavailable, and warn when a multi-replica backend has no peer endpoint at all. - Reject a plaintext peer endpoint when the gateway serves TLS. - Skip the sandbox watch poller on single-replica backends, where the local update bus already sees every write. - Rate-limit the peer owner cache sweep so an insert no longer scans the whole map under the lock. - Retry GET and HEAD on a pooled upstream the sandbox closed, instead of returning 502, and drop an emptied endpoint from the pool right away. - Document the gateway peer environment variables and the post-rollout ownership skew operators should expect. Signed-off-by: divesh <dgude@nvidia.com> * fix(server): harden HA supervisor ownership Signed-off-by: Drew Newberry <anewberry@nvidia.com> --------- Signed-off-by: Drew Newberry <anewberry@nvidia.com> Signed-off-by: divesh <dgude@nvidia.com> Co-authored-by: Drew Newberry <anewberry@nvidia.com> Co-authored-by: divesh <dgude@nvidia.com> Co-authored-by: Divesh Chowdary <47188680+FrostGod@users.noreply.github.com>	2026-09-21 21:22:12 +00:00
Seth Jennings	dbe36eaf85	fix(security): harden Vault credential transport (#3329) ... Reject non-loopback plaintext Vault endpoints, disable redirects, and support private CA bundles without weakening hostname verification. Update Helm configuration, documentation, operator skills, and regression coverage for OSSR-002. Signed-off-by: Seth Jennings <sjenning@redhat.com>	2026-09-15 19:56:09 +00:00
3 people	5548405fcb	feat(credentials): add provider credential storage drivers (#2437) ... * feat(credentials): add provider credential storage drivers Signed-off-by: Taylor Mutch <taylormutch@gmail.com> * fix(credentials): harden credential update handling Signed-off-by: Taylor Mutch <taylormutch@gmail.com> * fix(credentials): harden credential driver security, correctness, and performance Address review findings from the credential storage drivers PR: - Route additional_credentials through the driver on refresh to prevent silent data loss for multi-credential providers (e.g. AWS STS) - Clean up stored credential handles on CAS failure during refresh to prevent orphaned secrets in external backends - Enforce namespace validation in the Kubernetes Secrets driver to prevent cross-namespace credential access when allow_reference_namespace is not enabled - Cache Vault Kubernetes auth tokens with 80% TTL to avoid re-authenticating on every credential operation - Parallelize resolve_credentials in all three drivers using try_join_all for faster sandbox startup - Add existingSecret support for the KEK Secret to fix helm template/GitOps workflows where lookup returns empty and regenerates the key - Document RBAC blast radius for the Kubernetes Secrets credential driver and recommend a dedicated namespace Signed-off-by: Varsha Prasad <varshaprasad96@gmail.com> Signed-off-by: Varsha Prasad Narsing <varshaprasad96@gmail.com> * fix(credentials): add optimistic concurrency, fix thundering herd, parallelize operations Use resourceVersion optimistic concurrency with retry loop for K8s Secret ownership checks to prevent TOCTOU races. Switch Vault token cache from RwLock to Mutex with double-check pattern to prevent thundering herd on cache miss. Parallelize credential store and delete operations across independent keys using try_join_all. Signed-off-by: Varsha Prasad <varshaprasad96@gmail.com> Signed-off-by: Varsha Prasad Narsing <varshaprasad96@gmail.com> * fix(credentials): handle partial failures, add delete retry, consolidate cleanup Replace try_join_all with join_all in credential store/delete operations to handle partial failures — successfully-stored handles are cleaned up when another key fails. Add retry loop with conflict detection to db-credstore delete_credential, matching the K8s driver pattern. Consolidate 4 manual cleanup_pre_stored_provider_credentials call sites into a single error handler using an async block. Remove inconsistent .trim() from db-credstore validate_handle_owner. Signed-off-by: Varsha Prasad <varshaprasad96@gmail.com> Signed-off-by: Varsha Prasad Narsing <varshaprasad96@gmail.com> * fix(credentials): fix retry loop guard and remove unprotected validation Remove attempt-count guard from 409/Aborted match arms in retry loops so the post-loop Status::aborted error is reachable after exhausting retries. Previously, last-attempt conflicts fell through to the catch-all error arm, producing misleading Status::unavailable errors. Remove duplicate validation calls that ran after prepare_provider_credential_update but outside the cleanup-protected async block, which would leak pre-stored handles on failure. Signed-off-by: Varsha Prasad <varshaprasad96@gmail.com> Signed-off-by: Varsha Prasad Narsing <varshaprasad96@gmail.com> * fix(credentials): add workspace/provider UUID to credential backend paths Include workspace and provider ID in credential backend object paths to ensure cross-workspace uniqueness and prevent credential collision (GATOR-1806c9be-01). - Updated credential driver proto to include workspace and provider_id fields - Modified Vault driver to include workspace/provider_id in managed_secret_path - Modified Kubernetes Secrets driver to include workspace/provider_id in credential_owner_id and managed_secret_name - Updated all credential runtime calls to pass workspace/provider_id - Updated tests to use the new signatures This prevents two workspaces sharing the same external credential store from colliding on provider names, which was a critical security issue (CWE-639). * fix(credentials): preserve provider-level expiration for handle-backed credentials Compute effective expiration from both provider and driver values using the earliest non-zero timestamp and skip expired values before insertion (GATOR-1806c9be-02). - Modified resolve_provider_handles to check provider credential_expires_at_ms - Skip expired credentials during resolution instead of returning them - Use effective expiration (min of provider and driver) in resolution results - Fix inference.rs to preserve earliest expiration when merging This ensures handle-backed credentials respect the same expiration semantics as inline credentials. * fix(credentials): stage refresh changes under new handles before validation Stage credential replacements under new immutable handles instead of reusing existing handles to prevent overwriting committed values before validation/CAS (GATOR-1806c9be-03). - Stage credentials with empty existing_handles map to force new handle creation - Validate and CAS before the new values are committed to backend storage - Delete old handles only after successful CAS - On CAS failure, delete only the newly staged handles - This prevents CWE-362/CWE-367 race conditions where failed refreshes could still modify or delete the active credential The fix ensures that a rejected refresh cannot modify the backend object still referenced by the committed provider record. * fix(credentials): add timeouts to credential driver RPCs Apply configured timeouts to both startup capability negotiation and runtime RPCs to prevent indefinite hangs (GATOR-1806c9be-05). - Add DEFAULT_CREDENTIAL_DRIVER_RPC_TIMEOUT_SECS constant (30s) - Apply timeout to GetCapabilities during startup connection - Apply timeout to all runtime RPCs (store, delete, resolve) - Use tokio::time::timeout to bound the entire GetCapabilities operation during startup, not just the socket connection - Return contextual deadline errors on timeout This prevents a faulty or overloaded driver from hanging gateway operations indefinitely. * fix(credentials): fix test to use consistent workspace/provider identity The Kubernetes auth Vault resolve test was constructing a managed path with test-workspace/test-provider-id but sending default/prov-123 in the request, causing validation to reject the request (GATOR-18e32351-01). - Update test to use test-workspace and test-provider-id in the request to match the logical_path construction - This ensures the test exercises the intended code path and validates Kubernetes auth resolution properly The test now passes and correctly validates identity enforcement. * fix(credentials): use unique staging ID for refresh to avoid overwrites Stage refresh replacements under genuinely distinct immutable handles using a unique staging ID to prevent overwriting committed values (GATOR-1806c9be-03). - Generate a unique staging ID using UUID for each refresh operation - Use this staging ID when storing credentials instead of the real provider ID - Pass the same staging ID during cleanup on failure to delete only staged objects - This ensures deterministic paths (Vault) and object names (K8s) don't collide with the committed provider's credentials The fix prevents failed refreshes from silently replacing active credentials or breaking providers by deleting still-referenced backend objects. * fix(credentials): wrap credential driver RPCs in local timeouts Add local tokio::time::timeout wrappers around credential driver RPCs to bound non-compliant or stalled UDS peers (GATOR-1806c9be-05). - Wrap StoreCredential, DeleteCredential, and ResolveCredentials in local timeouts - Return contextual deadline_exceeded errors when timeouts occur - Keep existing gRPC timeout metadata for compliant implementations - GetCapabilities during startup was already wrapped in previous commit This ensures a faulty local driver cannot hang gateway operations indefinitely, even if it accepts the connection but never responds to the RPC. * fix(credentials): preserve ownership for staged refreshes Signed-off-by: Seth Jennings <sjenning@redhat.com> * fix(credentials): bound startup capability probe Signed-off-by: Seth Jennings <sjenning@redhat.com> * test(provider): authenticate credential handler requests Signed-off-by: Seth Jennings <sjenning@redhat.com> * fix(ci): grant actions read to credential driver e2e Signed-off-by: Seth Jennings <sjenning@redhat.com> --------- Signed-off-by: Taylor Mutch <taylormutch@gmail.com> Signed-off-by: Varsha Prasad <varshaprasad96@gmail.com> Signed-off-by: Varsha Prasad Narsing <varshaprasad96@gmail.com> Signed-off-by: Seth Jennings <sjenning@redhat.com> Co-authored-by: Taylor Mutch <taylormutch@gmail.com> Co-authored-by: Varsha Prasad Narsing <varshaprasad96@gmail.com>	2026-08-05 16:29:41 +00:00