Skip to main content

Poller Timing

A poller is a scheduled job that regularly checks a data source (like a SIEM, ticketing system, or email inbox) for new events. Each time it runs, it queries the data source for events within a specific time window — a "from" and "to" timestamp. Three settings control exactly how that window is calculated.

Quick Reference​

SettingWhat it controlsEffect on window
FrequencyHow often the poller runsSets the base window size
Window OffsetHow far back the window endsShifts the entire window backwards
Lookback RangeHow far back the window startsWidens the window at the start

Interactive Diagram​

Use the sliders below to see how each setting shapes the query window.

5m
The poller runs every 5m — the query window covers the 5m gap since the last run
0m
No offset — the window ends at the current time
0m
No lookback — window start is set by frequency only
Query from5m ago
Query tonow
Window size5 min
Query Windowquery from← pastnow5m frequency

Polling Frequency​

The most basic setting is how often the poller runs. With no other settings configured, the query window is simply:

  • From: the last time the poller ran
  • To: right now

The minimum supported frequency is 30 seconds.

The query window always covers exactly the gap since the last run. This means if the poller was down for an hour, the next run will automatically query from 1 hour ago to now — no events are skipped.

Window Offset​

Since polling windows don't intersect, any event that falls in a window the poller already queried will never be looked at again. This becomes a problem when a data source has an ingestion delay — an event might happen at time T, but not appear in the API until T+5 minutes. If the poller already queried up to T and moved on, that event will be missed permanently.

The window offset shifts the entire query window backwards by a set amount. The window keeps the same size, it just ends a little in the past instead of at the current moment, giving the data source time to catch up before the poller asks for data.

For example, with a 30-minute frequency and a 10-minute offset:

  • From: 40 minutes ago
  • To: 10 minutes ago

Lookback Range​

The lookback range extends the start of the query window further into the past, without moving the end. This makes the window wider.

For example, if a poller runs every 30 minutes with a 60-minute lookback:

  • From: 90 minutes ago
  • To: now

This is useful when events can arrive very late — for instance, an audit log entry might not appear in the API until 45 minutes after it happened. Without lookback, those entries would fall outside the query window entirely.

Window Offset vs. Lookback Range​

Window OffsetLookback Range
What movesBoth start and end shift backOnly start shifts back
Window sizeUnchangedWider
Response timeAlways delayed by the full offsetEvents ready early are picked up immediately
Use whenThe data source is never reliable before N minutes (e.g. a pipeline that always batches on a delay)The data source is usually ready quickly but can occasionally be late
Trade-offConsistent delay on every eventMay see the same event across multiple runs (deduplicated automatically)
note

A wider query window means the same event might be returned across multiple runs. The poller automatically deduplicates results, so re-fetching an event doesn't create duplicate investigations.