Transaction configuration

Open Configuration → Transactions to control when Glowroot stores detailed traces and how expensive optional collectors are. Settings apply per agent; on Central, pick the target agent in the dropdown before editing (rollup agents cannot be configured here — see Agents and rollups).

Slow threshold

Default is often 2000 ms (check your agent's defaults). When a transaction's wall time exceeds this threshold:

• The request becomes an active trace while it runs (live visibility where supported).
• On completion, the trace is stored and appears under the Traces tab (Transaction tabs).

There is no sampling percentage — every request slower than the threshold is traced. To reduce volume, raise the threshold or tighten per-transaction overrides rather than expecting probabilistic sampling.

Profiling workflow (temporary threshold)

Use the slow threshold as a capture lever when you are investigating a specific call (Re-engineering a slow call):

1. Note the current global threshold and any overrides.
2. Lower the threshold (or add a per-transaction override for the name you care about) so reproducing the issue stores a Trace.
3. Exercise the endpoint / job; open the Trace and follow How to read a Trace.
4. Restore the previous threshold (or remove the override) so routine traffic does not flood storage and the Traces list.

This is still not probabilistic sampling: every request above the (temporary) threshold is traced. Prefer a narrow per-transaction override over a very low global threshold on a busy JVM.

Profiling interval

Controls how often CPU stack samples are taken during traced (slow) requests. Lower interval = finer Profile tab detail and higher overhead. See Transaction tabs and How to read a Trace for how Profile relates to breakdown timers.

Thread stats

Optional collection of per-thread CPU and blocked/waiting time during traces. Useful for diagnosing contention; adds overhead on traced requests. Enable only when you need that detail.

Per-transaction overrides

You can override slow threshold (and related options) by transaction type or transaction name pattern. Use this for noisy endpoints that need a higher bar, or critical paths that need a lower one.

Match names to what Glowroot reports (often requestURI or plugin-derived names). Overrides are evaluated after the global default.

Central vs embedded

In Embedded, you configure the single local agent. In Central, each connected JVM has its own transaction config — changes do not propagate across agents automatically.

See also

• How to read a Trace · Re-engineering a slow call
• Transaction tabs — what each tab shows once traces exist
• Agents and rollups — leaf agent vs rollup in the Central dropdown
• What Glowroot does not do — no trace sampling %, no automatic retention tuning from this page