ZITADEL Docs
APIsBenchmarksv4.17.1

human password login benchmark of zitadel v4.17.1

Benchmark results of the v4.17.1 release of Zitadel.

Performance test results

MetricValue
Baselinenone
PurposeTest current performance
Test start2026-08-27 19:24:36 UTC
Test end2026-08-27 19:54:52 UTC
Test duration30min
Executed testhuman_password_login
k6 versionv2.1.0
VUs600
Client locationUS1
ZITADEL locationUS1
ZITADEL container specificationvCPU: 6
Memory: 6 Gi
Container min scale: 7
Container max scale: 7
ZITADEL Versionv4.17.1
ZITADEL SettingsEventstore autovacuum (new in v4.17): Enabled: true
VacuumThreshold: 1000000
AnalyzeThreshold: 1000000
Tuned for the ~4,000 events/s expected during the benchmark, per the tuning guide.
ZITADEL feature flagsEnabled: loginDefaultOrg (instance), oidcTokenExchange (system), enableBackChannelLogout (system), consoleUseV2UserApi (instance)
improvedPerformance (instance): IMPROVED_PERFORMANCE_PROJECT_GRANT, IMPROVED_PERFORMANCE_PROJECT, IMPROVED_PERFORMANCE_USER_GRANT, IMPROVED_PERFORMANCE_ORG_DOMAIN_VERIFIED
Disabled: loginV2 (explicitly false at instance), userSchema, debugOidcParentError, oidcSingleV1SessionTermination, permissionCheckV2
Captured from the instance before the run (sequence 7, set 2026-08-27 13:35:31 UTC); the same capture covers manipulate_user at 13:41 and the rest of the sweep from 14:56, so all eleven targets ran the same configuration.
This restores like-for-like comparison with the v4 pages, which ran five improvedPerformance options. The fifth, IMPROVED_PERFORMANCE_ORG_BY_ID, is reserved in the proto and no longer exists.
Databasetype: psql
version: v17.4
Database locationUS1
Database specificationvCPU: 8
memory: 32Gib
ZITADEL metrics during testCPU: p50 48%, p99 60% (peak 64%)
Memory: p50 6%, peak 16%
Instances: 7 (constant), no container startups recorded
Server-side: 338 req/s, latency p50 207ms / p95 4.9s / p99 32.9s
Observed errors9,227 of 683,256 requests failed (1.35%). Corrected by hand from the generated row, which reported these as one population; the status breakdown shows two with different causes and different shapes. (1) 8,178 request timeouts (1.20%), sustained across the whole run — not a burst: they run from 19:26 to 19:54 at a steady 125–360 per minute, with no onset and no recovery. This is ZITADEL under 600 VUs on this target and is the number to read. (2) 1,044 HTTP 503 platform errors, confined to 19:37–19:44Service error -27, an HTML page from the Google frontend rather than a ZITADEL JSON error, in the same bounded-burst shape seen on manipulate_user at 13:43 the same day and pending the same GCP post-mortem (see benchmark/postmortems/2026-08-27-503-burst.md). Plus 4x HTTP 502 and 1x connection reset, scattered singles. 9,093 of 1,007,953 checks failed (0.90%): login name status ok (8,345), authorize status ok (333), token status ok (138), password status ok (137), userinfo status ok (119), password callback (21). The failed checks track the timeouts rather than the 503 burst.
Top 3 most expensive database queries1: Projection handler advisory lock, one per projection per instance — lock wait, not query work (17,902 s)
2: Eventstore read barrier: advisory lock + unlock on events2lock wait (8,920 s)
3: Query events by instance_id, aggregate_types, event_types, position (paged) (3,810 s)
Ranked by total accumulated execution time summed across concurrent sessions (Cloud SQL Query Insights) — not CPU time; it can exceed the wall-clock window.
Database metrics during testCPU: 87% median, 90% peak
Memory: 40%
Connections: 46
Transactions: 12,351/s
IOPS: read 0, write 711
CPU-saturated — the bottleneck for this target.
k6 Iterations per second49
k6 outputoutput
flowchart outcomePotential issues. Not resource-bound: database CPU 87%, ZITADEL CPU 48%, yet 86% of the top-5 query time is advisory lock wait — predominantly the projection handler lock, which serialises this workload. Additional vCPU will not help. Logged as an issue rather than blocking further scale testing. The run also contains a 1,044-request platform 503 burst (19:37-19:44) under separate investigation in benchmark/postmortems/2026-08-27-503-burst.md; it does not affect this verdict.

Endpoint latencies

k6 output

 TOTAL RESULTS

    checks_total.......: 1007953 542.050139/s
    checks_succeeded...: 99.09%  998860 out of 1007953
    checks_failed......: 0.90%   9093 out of 1007953

 user defined
 authorize status ok
  99% 90339 / 333
 login name status ok
  90% 81994 / 8345
 login shows password page
 password status ok
  99% 81702 / 137
 password callback
  99% 81681 / 21
 code set
 token status ok
  99% 81543 / 138
 access token created
 id token created
 info created
 org created
 create user is status ok
 openid configuration
 userinfo status ok
  99% 81423 / 119
 org removed

    CUSTOM
    human_password_login_duration........: min=652ms    avg=7.16s    max=1m9s    p(50)=4.76s    p(95)=16.62s   p(99)=1m2s
    login_ui_enter_login_name_duration...: min=65.11ms  avg=3.4s     max=1m0s    p(50)=400.69ms p(95)=11.5s    p(99)=59.35s
    login_ui_enter_password_duration.....: min=597.54µs avg=25.21ms  max=8.43s   p(50)=2.77ms   p(95)=6.61ms   p(99)=459.33ms
    login_ui_init_login_duration.........: min=29.53ms  avg=505.93ms max=6.4s    p(50)=94.64ms  p(95)=1.77s    p(99)=2.52s
    login_ui_token_duration..............: min=41.07ms  avg=651.58ms max=7.5s    p(50)=161ms    p(95)=2.2s     p(99)=3.04s
    oidc_user_info_duration..............: min=11.81ms  avg=103.78ms max=2.79s   p(50)=31.68ms  p(95)=334.09ms p(99)=489.53ms
    org_create_org_duration..............: min=47.19ms  avg=47.19ms  max=47.19ms p(50)=47.19ms  p(95)=47.19ms  p(99)=47.19ms
    user_create_human_duration...........: min=349.88ms avg=3.46s    max=6.17s   p(50)=3.42s    p(95)=5.8s     p(99)=6.14s

    HTTP
    http_req_duration....................: min=597.54µs avg=1.58s    max=1m0s    p(50)=146.67ms p(95)=4.08s    p(99)=1m0s
      { expected_response:true }.........: min=1.3ms    avg=879.85ms max=1m0s    p(50)=141.02ms p(95)=3.46s    p(99)=9.21s
    http_req_failed......................: 1.35%  9227 out of 683256
    http_reqs............................: 683256 367.436785/s

    EXECUTION
    iteration_duration...................: min=652.1ms  avg=11.96s   max=1m9s    p(50)=5.2s     p(95)=1m0s     p(99)=1m2s
    iterations...........................: 90671  48.760436/s
    vus..................................: 0      min=0              max=600
    vus_max..............................: 600    min=600            max=600

    NETWORK
    data_received........................: 1.5 GB 810 kB/s
    data_sent............................: 325 MB 175 kB/s

running (30m59.5s), 000/600 VUs, 90671 complete and 0 interrupted iterations
default [ 100% ] 600 VUs  30m0s

Was this page helpful?

On this page