The last unanswered action, and the only one whose output is not the input with edits -- it emits a record that never existed, which is why it was deferred twice. It emits the window's first record unchanged, tagged with cairnobs.aggregated, cairnobs.count, and the observed window bounds. That follows the convention the agent already uses for heartbeat and host-metrics records rather than inventing a second synthetic-record mechanism, and keeping the first record intact means a reader sees a real example of what was collapsed instead of an invented summary. It tags even when the count is one. Emitting a bare record there would be tidier and would make cairnobs.count present only sometimes, so summing it silently breaks on quiet windows. window_last is the last record that actually contributed, never window_start + window_ms, because a window flushed early must not claim an end that never happened. Specifying it surfaced a problem the other nine actions do not have. Windows are measured on record time, so a window can only be closed by a later record arriving. suppress_duplicates never has anything pending; aggregate_count holds state, so a matching stream that goes quiet leaves its aggregate unemitted indefinitely -- data loss dressed as latency. Emission therefore has a second trigger, end of stream, which the corpus defines as an implicit flush after the last input and which production gets from the batch flush. The cost is stated rather than hidden: window_ms becomes a maximum, not a guarantee, and one burst can produce more than one aggregate. And it has a consequence nobody should meet in production first: stats count undercounts aggregated data silently, so every panel and alert counting rows changes meaning the moment a rule aggregates the data behind it. Nothing here fixes that. The correct idiom is summing cairnobs.count; teaching the query layer to do it automatically is a Phase 2 change to the IR, recorded as the open question this decision leaves in its place rather than quietly inherited. Six cases added, corpus at 45. The validator's unspecified-action guard stays in place with an empty set, still rejecting anything added to it. Signed-off-by: John Coffey <[email protected]>
processing
The Phase 8 processing rule language: its specification, and the conformance suite that is that specification.
Status: the suite exists; the implementations do not. Nothing in Phase 8 is built. See
/docs/phase-8-processing-design.mdfor the design and its open questions.
Why this is a top-level directory
Rules run in two places, in two languages: the Rust agent
(/agent) and the Go ingest tier (/ingest). Neither owns the
definition. This is the same shape as /proto — a language-neutral
contract that several modules consume and none of them is the source of
truth for.
Why the cases are the specification
Two hand-written implementations of one language will diverge. Not maybe, eventually. The only thing that reliably prevents it is a shared corpus both must satisfy, written down before either exists.
So the prose below is a summary of conformance/cases/, not the other
way round. When they disagree, the cases win, and the prose is the bug.
Running it
Nothing executes the cases yet, because neither implementation exists. What runs today is a structural check that every case is well-formed — valid JSON, known actions, no unknown keys, declared fields that exist:
python3 processing/conformance/validate.py
That is worth having on its own: it stops the corpus rotting between now and the first implementation, and it fails loudly if someone adds a case using an action the spec does not define.
Each implementation is expected to add a test that walks
conformance/cases/*.json, feeds inputs through rules, and asserts
the emitted records equal expect. Those runners are part of building
each side, not part of this directory.
Case format
One JSON object per file:
{
"name": "drop_discards_the_record",
"description": "A matched drop emits nothing at all.",
"rules": [
{
"match": [{"field": "message", "op": "eq", "value": "noise"}],
"actions": [{"action": "drop"}]
}
],
"inputs": [
{"timestamp_unix_nano": 1000, "host": "h1", "service": "s1",
"severity": "SEVERITY_INFO", "message": "noise", "attributes": {}}
],
"expect": []
}
inputs is always a list, even for a single record, because windowed
actions need a sequence. expect is the records emitted, in order.
Records
Mirrors LogRecord in /proto/sentry/logs/v1/logs.proto, snake_case,
with severity as the enum's name string.
record_id is deliberately absent from every case. The agent never sets
it, and whether an ingest-side rule can see one depends on where in the
ingest path rules run — an open question. No case may depend on it, so
that decision stays free.
Matching
A rule matches when every clause in match holds. An empty match
matches every record.
op |
Meaning |
|---|---|
eq / ne |
exact string equality |
contains |
substring |
prefix / suffix |
string boundary |
regex |
linear-time regex, unanchored |
exists / not_exists |
field present (no value) |
Addressable fields: message, host, service, severity, and
attributes.<key>. A field that does not exist compares as absent, not
as empty string — eq "" does not match a missing attribute.
Actions
Applied in order. A drop ends processing for that record immediately;
no later action or rule runs.
| Action | Parameters |
|---|---|
drop |
— |
drop_fields |
fields |
keep_fields |
fields (attributes only; never removes top-level fields) |
rename |
from, to |
derive |
field, and one of value / from_field |
mask |
field, pattern, replacement |
parse_json |
field (default message), optional prefix |
parse_regex |
field, pattern (named captures become attributes) |
sample |
keep_one_in |
suppress_duplicates |
window_ms, optional key_fields |
aggregate_count |
window_ms, optional key_fields |
Rules are evaluated in the order given. Every matching rule's actions apply, to the record as left by the rule before it.
Regex: only what both engines support
The agent uses regex-lite and ingest uses the full regex crate — a
size decision, since the agent ships as a small static musl binary and
the full crate is a megabyte-plus (see
phase-8-processing-design.md).
Cases may only use syntax regex-lite accepts. A case the agent
cannot run is not a conformance case. In practice that means quantifiers,
character classes, alternation and named captures are fine, while
Unicode-aware classes and the richer Perl classes are not.
Both engines are linear-time automata with no catastrophic backtracking, and both resolve alternation leftmost-first. The second is pinned by a case rather than trusted, because it is exactly the kind of semantic two independent implementations can differ on silently.
Two determinism decisions the suite forces
Conformance testing cannot assert on nondeterminism, so two things that are usually left vague are pinned here:
sample is counter-based, not random. keep_one_in: 3 keeps the
first record and every third thereafter, per rule, per process. Random
sampling is statistically nicer and untestable; a counter is testable and
close enough at volume.
Windows are measured on record timestamps, not wall-clock. A
suppress_duplicates window of 5000ms compares
timestamp_unix_nano values, so replaying the same input always gives
the same output regardless of how fast the test runs. This also means the
behaviour is correct under backfill, which wall-clock windows are not.
aggregate_count emits a record that never existed
The only action whose output is not the input with edits, so its shape is worth stating here too. It emits the window's first record, unchanged, plus four attributes:
| attribute | value |
|---|---|
cairnobs.aggregated |
"true" |
cairnobs.count |
records collapsed, as a string |
cairnobs.window_start_unix_nano |
first record's timestamp |
cairnobs.window_last_unix_nano |
last contributing record's timestamp |
This follows the convention the agent already uses for heartbeat and
host-metrics records: an ordinary record distinguished by a
cairnobs.* attribute, rather than a second synthetic-record mechanism.
Three behaviours the cases pin, each of which is easy to get wrong:
- It tags even when the count is one. Otherwise
cairnobs.countis present only sometimes, and summing it silently breaks on quiet windows. window_lastis observed, neverwindow_start + window_ms. A window flushed early must not claim an end that never happened.- A pending window flushes at end of stream. Record-time windows can
only be closed by a later record, so without this a burst that stops
leaves its aggregate unemitted indefinitely. In production the batch
flush provides the same trigger, which makes
window_msa maximum rather than a guarantee.
stats count undercounts aggregated data. The correct idiom is
summing cairnobs.count. Whether the query layer should do that
automatically is a Phase 2 question, recorded in
phase-8-processing-design.md.
Adding a case
One behaviour per file, named for the behaviour rather than the action. A case is added for every bug found in either implementation — that is the mechanism that keeps the two from drifting, and it only works if the case lands with the fix.