exd telemetry version-skew
Eval counts grouped by manifest_version. How spread out are your SDK clients across versions of the manifest?
Synopsis
exd telemetry version-skew
[--flag <key>]...
[--namespace <slug>] [--environment <env>]
[--since <duration> | --from <ts> --to <ts>]
[--source <uri>]...
[--format human|json]
[--limit <n>]
[--fail-on-error]
See common flags for shared options.
Description
Records carry the manifest_version the SDK was on when it evaluated. This command counts records per version. A healthy fleet runs nearly everyone on the current version; a wide spread means some SDKs are stuck on stale closures.
Use cases
-
Diagnose SRM failures.
srmfiresT001; checkversion-skewto see whether stale SDKs are serving the old split. -
Verify rollouts have actually propagated. Push v13, wait for the polling/SSE refresh window, run
version-skew— if v12 is still significant, refresh isn't working:exd telemetry version-skew --since 1h -
Audit before retiring a flag. Old SDK versions may still reference a flag the current manifest deleted. Combine with the version skew to estimate impact.
-
CI gate on rollout propagation. Fail the deploy job if version skew exceeds the threshold:
exd telemetry version-skew --since 30m --fail-on-error
Subcommand-specific flags
None beyond the common-flag set.
Result fields (--format json)
result.versions is an array, sorted descending by count:
| Field | Type | Notes |
|---|---|---|
manifest_version | integer | Version. |
count | integer | Evaluation records on that version. |
percent | float | Share of all records in the window. |
Diagnostics
T004manifest-version skew (when the spread exceedsversion_skew.max_versionsor then-th-version share exceedsversion_skew.tail_pct)
Example
$ exd telemetry version-skew --since 24h
Eval counts by manifest_version (last 24h)
manifest_version count percent
13 1,184,422 86.4%
12 178,109 13.0% ← T004
11 8,221 0.6%
T004 warning: manifest-version skew — 13.6% of traffic is on a non-current version.
Likely causes: SDKs not refreshing, SSE channel down, or cache staleness.
Investigate by environment: --environment prod, --environment staging.
Exit codes
See telemetry exit codes.