feat: Add SCIM scenarios; Fix findings - #1136
Merged
Merged
Conversation
|
🦢 Load Test Results Goose Attack ReportPlan Overview
Request Metrics
Response Time Metrics
Status Code Metrics
Transaction Metrics
Scenario Metrics
Error Metrics
|
|
🧪 Tempest Identity Compatibility Results (advisory, non-blocking) rust Failed test IDs |
|
| Project | keystone |
| Branch | claude/loadtest-expansion-stress-u8i0j4 |
| Testbed | ubuntu-latest |
Click to view all benchmark results
| Benchmark | Latency | Benchmark Result nanoseconds (ns) (Result Δ%) | Upper Boundary nanoseconds (ns) (Limit %) |
|---|---|---|---|
| Command_Serde/apply/remove | 📈 view plot 🚷 view threshold | 118,880.00 ns(-62.77%)Baseline: 319,272.06 ns | 1,634,101.99 ns (7.27%) |
| Command_Serde/apply/set | 📈 view plot 🚷 view threshold | 124,040.00 ns(-64.49%)Baseline: 349,281.86 ns | 1,679,671.23 ns (7.38%) |
| Command_Serde/pack/delete | 📈 view plot 🚷 view threshold | 123.94 ns(+4.40%)Baseline: 118.71 ns | 157.53 ns (78.68%) |
| Command_Serde/pack/delete_index | 📈 view plot 🚷 view threshold | 113.57 ns(+8.01%)Baseline: 105.15 ns | 139.12 ns (81.63%) |
| Command_Serde/pack/set | 📈 view plot 🚷 view threshold | 201.57 ns(+6.39%)Baseline: 189.47 ns | 248.75 ns (81.03%) |
| Command_Serde/pack/set_index | 📈 view plot 🚷 view threshold | 109.56 ns(+4.18%)Baseline: 105.17 ns | 139.18 ns (78.72%) |
| Command_Serde/unpack/delete | 📈 view plot 🚷 view threshold | 190.20 ns(+3.03%)Baseline: 184.61 ns | 267.96 ns (70.98%) |
| Command_Serde/unpack/delete_index | 📈 view plot 🚷 view threshold | 175.28 ns(+13.12%)Baseline: 154.95 ns | 218.78 ns (80.12%) |
| Command_Serde/unpack/set | 📈 view plot 🚷 view threshold | 288.31 ns(+12.36%)Baseline: 256.60 ns | 349.57 ns (82.48%) |
| Command_Serde/unpack/set_index | 📈 view plot 🚷 view threshold | 159.99 ns(+5.66%)Baseline: 151.42 ns | 212.26 ns (75.37%) |
| Payload_encryption/pack/remove_cmd | 📈 view plot 🚷 view threshold | 119.84 ns(+5.59%)Baseline: 113.50 ns | 151.03 ns (79.35%) |
| Payload_encryption/pack/set_cmd | 📈 view plot 🚷 view threshold | 196.94 ns(+0.80%)Baseline: 195.39 ns | 263.37 ns (74.78%) |
| Payload_encryption/unpack/remove_cmd | 📈 view plot 🚷 view threshold | 199.16 ns(+3.12%)Baseline: 193.13 ns | 281.45 ns (70.76%) |
| Payload_encryption/unpack/set_cmd | 📈 view plot 🚷 view threshold | 314.84 ns(+16.56%)Baseline: 270.11 ns | 374.42 ns (84.09%) |
| Raft_1Node_Latency/prefix/1node | 📈 view plot 🚷 view threshold | 3,427,200.00 ns(+27.68%)Baseline: 2,684,171.25 ns | 6,171,231.98 ns (55.54%) |
| Raft_1Node_Latency/read/1node | 📈 view plot 🚷 view threshold | 42,163.00 ns(+10.75%)Baseline: 38,068.81 ns | 53,319.60 ns (79.08%) |
| Raft_1Node_Latency/remove/1node | 📈 view plot 🚷 view threshold | 351,450.00 ns(-57.04%)Baseline: 818,054.53 ns | 4,098,046.95 ns (8.58%) |
| Raft_1Node_Latency/write/1node | 📈 view plot 🚷 view threshold | 362,060.00 ns(-62.72%)Baseline: 971,300.16 ns | 5,492,419.70 ns (6.59%) |
| build_snapshot/default | 📈 view plot 🚷 view threshold | 107,910.00 ns(-13.77%)Baseline: 125,140.23 ns | 219,356.50 ns (49.19%) |
| fernet token/project | 📈 view plot 🚷 view threshold | 1,457.70 ns(+2.23%)Baseline: 1,425.89 ns | 1,811.00 ns (80.49%) |
| get_data_keyspace | 📈 view plot 🚷 view threshold | 0.32 ns(+1.91%)Baseline: 0.32 ns | 0.39 ns (82.85%) |
| get_db | 📈 view plot 🚷 view threshold | 0.32 ns(+1.56%)Baseline: 0.32 ns | 0.39 ns (82.69%) |
| get_fernet_token_timestamp/project | 📈 view plot 🚷 view threshold | 162.96 ns(+16.03%)Baseline: 140.44 ns | 179.00 ns (91.04%) |
| get_keyspace | 📈 view plot 🚷 view threshold | 4.32 ns(-29.96%)Baseline: 6.16 ns | 14.58 ns (29.61%) |
gtema
force-pushed
the
claude/loadtest-expansion-stress-u8i0j4
branch
3 times, most recently
from
August 3, 2026 14:20
10007b9 to
b5d0c3d
Compare
Add ScimRealmRead and ScimSync loadtest scenarios covering raft-backed federation/SCIM (ADR 0021, 0024): realm CRUD reads/updates via admin token, and the real IdP-connector lifecycle over /SCIM/v2 with a per-VU API Key bearer (provision, reconcile, attribute drift, offboard). Seed setup is now idempotent against a pre-existing SCIM realm/IdP (no delete endpoint exists, so an interrupted prior run leaves them behind; 409 now reuses them instead of aborting seed). While chasing intermittent 401s surfaced by ScimSync under load, found and fixed two real bugs in the raft-backed auth path: - Timing oracle in ApiKeyAuth: the IP-not-allowed rejection branch returned immediately while every other rejection branch (not-found, inactive, wrong-secret) burns an Argon2id hash first, letting an attacker distinguish a valid-but-IP-blocked lookup_hash from a nonexistent one by response latency (ADR 0021 Invariant 7). Now burns the same dummy-hash cost. - Two linearizability gaps in the ReadIndex forward-to-leader path (get_by_key/prefix/prefix_index): the leader-side gRPC handlers served forwarded reads without re-confirming current leadership, and follower-side callers silently fell back to an unguarded local read whenever the forward RPC failed. Both fixed. fix(scim): map raft revision conflict to 409 instead of 500 The RaftBackend only matched StoreError::Conflict to convert ApiStoreError::Conflict to ScimRealmProviderError::Conflict. However, the the Conflict arm never matched -- concurrent PUT /v4/scim_realms returned 500 Internal Server Error instead of 409 Conflict. Also restructure update_impl error handling to properly separate Conflict, NotFound, and raft storage errors. fix(storage): Fall back to local read when forward fails When a forwarded gRPC read fails (e.g., a removed node has stale leader cache pointing to an old leader), fall back to local read from the FjallDB instead of hard-failing. This allows removed nodes to serve their last committed state. Also add test_prefix_index_leader_race_concurrent regression test for GitHub #1135 to verify prefix_index works correctly under concurrent load on the leader. Signed-off-by: Artem Goncharov <artem.goncharov@gmail.com>
gtema
force-pushed
the
claude/loadtest-expansion-stress-u8i0j4
branch
from
August 3, 2026 17:22
b5d0c3d to
4acc50e
Compare
Open
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add ScimRealmRead and ScimSync loadtest scenarios covering raft-backed
federation/SCIM (ADR 0021, 0024): realm CRUD reads/updates via admin
token, and the real IdP-connector lifecycle over /SCIM/v2 with a
per-VU API Key bearer (provision, reconcile, attribute drift,
offboard). Seed setup is now idempotent against a pre-existing SCIM
realm/IdP (no delete endpoint exists, so an interrupted prior run
leaves them behind; 409 now reuses them instead of aborting seed).
While chasing intermittent 401s surfaced by ScimSync under load,
found and fixed two real bugs in the raft-backed auth path:
Timing oracle in ApiKeyAuth: the IP-not-allowed rejection branch
returned immediately while every other rejection branch (not-found,
inactive, wrong-secret) burns an Argon2id hash first, letting an
attacker distinguish a valid-but-IP-blocked lookup_hash from a
nonexistent one by response latency (ADR 0021 Invariant 7). Now
burns the same dummy-hash cost.
Two linearizability gaps in the ReadIndex forward-to-leader path
(get_by_key/prefix/prefix_index): the leader-side gRPC handlers
served forwarded reads without re-confirming current leadership,
and follower-side callers silently fell back to an unguarded local
read whenever the forward RPC failed. Both fixed.
Signed-off-by: gtema artem.goncharov@gmail.com