Filebeat cursor cache: can receiver startup order change the state index?
Input identity and cached-store ownership must be checked separately.
Store ownership beyond success logs
Input identity and cached-store ownership must be checked separately. Each input must read its own state index regardless of restart order. The Truthound reference comes from the deployment manifest, not the index accidentally chosen by that run; two successful readers of one shared index do not establish isolation. [1, 2, 3]
The index chosen by initialization
Beats #53034 describes receivers sharing path.data and input type also sharing a cached cursor store bound by the first initializer. Reversing startup order can select another state index. Its evidence is a main-commit code path and reproduction outline, not a quantified inventory of stale cursors or duplicated production records. [1]
Change only startup order
Reverse two receivers’ startup order and compare each input’s actual state index. Hold resume when an input reads another input’s stale state. Record backend, input type and path.data while varying startup order. Correct index ownership still needs separate cursor-revision and committed-sink checks. [1, 3]
Where B must read
In the illustrative deployment, input B owns agentless-state-B. That actual index meets the ownership condition. agentless-state-A fails even if it happens to contain a cursor for B. The expected index is defined independently in the manifest, not copied from the cache result. [1, 3]
Compare the manifest with observation
Truthound Core Python SDK 3.3.2 can compare prepared owned_state_index and actual_state_index through ColumnPairValidator with relationship="==". The operator observes each input’s selected index and supplies non-missing names. The SDK does not split the store cache or rewind cursors, and matching names do not establish commit atomicity. [1, 2, 3]
Preserve cursors before a workaround
Consider separate path.data isolation or a supported correction, then reconcile historical events by source key after restart. Different backend or input type changes the reported shared-cache condition. Preserve original cursor state before applying a workaround so the recovery reference is not lost. [1, 3]
Application conditions
Record backend, input type, path.data and actual index; define expected ownership in the deployment manifest.
Alternatives
Test a supported correction or path.data isolation and reconcile reingestion by source keys.
Limitations
The SDK does not split cache or edit cursors; correct index names do not establish cursor freshness or commit atomicity.
Evidence
- Event checked
- Subject checked
- Modified
