sequence
Find entities whose events match a list of conditions in order, optionally within a time limit.
Use sequence to find users, hosts, or IPs whose events match your conditions in order. You can set a maximum span for patterns such as failed logons followed by success, or process launch followed by an outbound connection.
Syntax
... | sequence by <field>[, <field> ...] [maxspan=<duration>] [fields(<field>, ...)] [<condition 1>] [<condition 2>] ...The parts must come in this order: by, maxspan, fields(...), then the conditions.
| Argument | Default | Description |
|---|---|---|
by | required | Fields that identify the entity. Events where a by field is empty are left out. |
maxspan | no limit | Longest time from the first step to the last, for example 5m, 1h, 24h. |
fields(...) | none | Extra fields to capture from each step's event. |
[condition] | at least one | A filter in square brackets. Conditions use the where expression syntax (NOT > AND > OR): comparisons, IN, LIKE, regex (field=/pattern/), IN CIDR(...) and functions. |
maxgap= is accepted for compatibility and ignored.
Output fields
One row per entity that completed every step:
| Field | Description |
|---|---|
by fields | The entity |
step<N>_time | Time of the event that matched step N |
step<N>_event_id | That event's metadata.uid, for drill-down |
step<N>_<field> | Fields named in step N's condition and in fields(...), captured from that event. Dots become underscores: process.cmd_line → step1_process_cmd_line. |
sequence_duration_seconds | Seconds from step 1 to the last step |
maxspan_seconds | The maxspan value, when set |
sequence_count | Number of completed sequences for this entity |
Rows are ordered by step1_time, newest first.
How it works
- The reported step columns describe one run that completes every condition in order and satisfies
maxspan, when set. Ordering uses event timestamps at millisecond resolution. - Entities that stop partway are left out. Use funnel to see where they stop.
- The base search limits which events are considered. Filter it to the log sources and classes your conditions need.
Examples
Failed logons followed by a successful logon for the same account within 5 minutes:
class_uid=3002 | sequence by user.name maxspan=5m [status_id=2] [status_id=1]The same, per account and source IP, keeping the target host:
class_uid=3002 | sequence by user.name, src_endpoint.ip maxspan=10m fields(device.hostname) [status_id=2] [status_id=1]PowerShell started, then PowerShell made an outbound connection from the same host within 10 minutes:
metadata.log_source=windows_sysmon | sequence by device.hostname maxspan=10m fields(process.cmd_line, dst_endpoint.ip) [class_uid=1007 process.name="powershell.exe"] [class_uid=4001 actor.process.name="powershell.exe"]Download from a file-sharing site, then a request for an executable from the same client:
class_uid=4002 | sequence by src_endpoint.ip maxspan=10m fields(http_request.url.url_string) [http_request.url.hostname=/file[-_]?(share|drop|transfer)/] [http_request.url.path=/\.(exe|scr|hta|ps1|msi)$/]A process writes a file under a temp folder, then a process starts from that host within 5 minutes:
metadata.log_source=windows_sysmon | sequence by device.hostname maxspan=5m fields(file.path, process.cmd_line) [class_uid=1001 file.path LIKE "%\\temp\\%"] [class_uid=1007]Failure-then-success patterns, slowest first:
class_uid=3002 | sequence by user.name maxspan=5m [status_id=2] [status_id=1] | table user.name, step1_time, step2_time, sequence_duration_seconds | sort -sequence_duration_secondsNotes
- In a scheduled detection rule, set the rule's lookback to at least
maxspan. Otherwise the first steps of a pattern can fall outside the window. See Detections. - Shorter
maxspanvalues and narrower base searches run faster.
What's next
- funnel: how many entities reach each step, and where they drop off
- transaction: group events by key without ordered conditions
- where: condition syntax
- lateral: trace movement between hosts