vinzenz/hyperhive
vinzenz/hyperhive
Watch
0
0
Fork
You've already forked hyperhive
 0
Code Activity

swarm: keep an agent's terminal rows and replay them on attach

Browse source 
swarm-controller now creates `term-main-<agent>` for every live agent, in
the same one-minute pass and with the same config style as
`term-sub-<agent>` (#4897): subjects `$SWARM.term.<agent>` only, max_age
24h, max_bytes 64 MiB, discard Old, max_message_size from the queue's
max_payload. An existing stream is opened as it is. The agent keeps its
plain publish on `$SWARM.term.<agent>` and gets no JetStream subject.

`GET /api/agents/{name}/term/stream` creates an ephemeral push consumer on
that stream starting 100 rows back, so a swarm-ui terminal opens with
recent history and then follows new rows, in order, from one
subscription. Each SSE event carries the row's stream sequence as its id,
and a reconnect's Last-Event-ID resumes after it. Without the stream the
route falls back to the live core subscription it used before.

The stream is `term-main-<agent>`, not `term-<agent>`: an agent may be
named `sub-x`, whose `term-<agent>` would be agent `x`'s subagent stream.
Its one literal subject cannot match `$SWARM.term.<agent>.sub.>`, which
needs more tokens, so the two streams never overlap.

The reader gains `$JS.API.CONSUMER.CREATE.*.*.$SWARM.term.*`, the
extended create form whose last tokens are the consumer's filter. Checked
against nats-server 2.14.4, the trailing `*` also admits a literal `*` or
`>` filter, so a consumer under it reaches any terminal row, main or
subagent; those are rows the reader already subscribes to and reads with
`STREAM.MSG.GET.*`.

Closes #4826
This commit is contained in:
atlas 2026-10-11 12:35:31 +02:00
commit ae9ca4331e
12 changed files with 533 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

281
swarm-queue-client/src/agent_term.rs Normal file
View file

@ -0,0 +1,281 @@
//! An agent's own terminal: the stream that keeps its rows, and the consumer
//! a reader replays them through.
//!
//! The agent publishes its classified terminal rows on
//! [`crate::term_subject`] with a plain core publish, under its own queue
//! credential, which may publish there and nothing else of these streams.
//! `swarm-controller` creates one stream per agent with the fixed config
//! [`crate::agent_term::open_or_create`] spells, so the queue keeps the rows a
//! reader who attaches later replays.
//!
//! The stream captures exactly one literal subject, `$SWARM.term.<agent>`.
//! [`crate::subagent_term`]'s stream captures `$SWARM.term.<agent>.sub.>`,
//! which only matches subjects with at least two more tokens, so no subject
//! is in both and the two streams never overlap. The `JetStream` server refuses
//! to create a stream whose subjects overlap another's.
//!
//! The names live here because three crates must agree on them: the
//! publisher in `hive-agent`, `swarm-controller`, and the reader's consumer
//! grant in `swarm-nats-auth`.
/// The token between the prefix and the agent in [`stream_name`]. It differs
/// from `term-sub-` at the same position, so no agent name makes this
/// stream's name equal another agent's subagent stream.
const STREAM_KIND: &str = "main";
/// How long the stream keeps a row. The same window as the subagent stream.
pub const MAX_AGE: std::time::Duration = std::time::Duration::from_hours(24);
/// How large one agent's stream may grow before the discard policy reclaims
/// space. The same cap as the subagent stream, for the same reason: it
/// bounds the disk one agent can claim.
#[cfg(feature = "agent-term")]
const MAX_BYTES: i64 = 64 * 1024 * 1024;
/// How many of the newest rows a reader that attaches fresh is replayed
/// before the live tail. The agent's own web UI loads the same number of
/// events as its first history page.
pub const REPLAY_ROWS: u64 = 100;
/// The stream holding `agent`'s terminal rows: `term-main-<agent>`.
#[must_use]
pub fn stream_name(agent: &str) -> String {
format!("term-{STREAM_KIND}-{agent}")
}
/// The one subject [`stream_name`]'s stream captures.
#[must_use]
pub fn stream_subject(agent: &str) -> String {
crate::term_subject(agent)
}
/// The reader's grant to create a replay consumer: on any stream, under any
/// consumer name, in the extended create form whose last tokens are the
/// consumer's filter.
///
/// The server refuses a create whose payload filter differs from those
/// tokens, so a consumer made under this grant delivers only rows on
/// subjects below `$SWARM.term`. The grant's last `*` also matches a literal
/// `*` or `>` token, so besides one agent's [`crate::term_subject`] it admits
/// `$SWARM.term.*` and `$SWARM.term.>`, which take in every agent's and every
/// subagent's terminal. The bare `CONSUMER.CREATE.<stream>` form, whose
/// filter rides only in the payload, is not covered.
#[must_use]
pub fn consumer_create_grant() -> String {
format!(
"$JS.API.CONSUMER.CREATE.*.*.{}",
crate::term_subject_wildcard()
)
}
/// The first stream sequence to replay from.
///
/// `resume_after` is the last sequence the reader already holds, from a
/// reconnecting browser's `Last-Event-ID`; the replay continues right after
/// it. A fresh reader gets the newest [`REPLAY_ROWS`] below `last_sequence`,
/// the stream's newest row when it was read. A sequence the stream has since
/// discarded starts the consumer at the oldest row it still has.
#[must_use]
pub fn replay_start(last_sequence: u64, resume_after: Option<u64>) -> u64 {
match resume_after {
Some(seen) => seen.saturating_add(1),
None => last_sequence.saturating_sub(REPLAY_ROWS - 1).max(1),
}
}
/// The config [`open_or_create`] creates `agent`'s stream with.
///
/// `max_message_size` tracks the queue's live `max_payload`, as the subagent
/// stream's does, so the stream never rejects a row the queue accepted.
#[cfg(feature = "agent-term")]
fn stream_config(agent: &str, max_message_size: i32) -> async_nats::jetstream::stream::Config {
async_nats::jetstream::stream::Config {
name: stream_name(agent),
description: Some(format!("Terminal rows of {agent}")),
subjects: vec![stream_subject(agent)],
max_age: MAX_AGE,
max_bytes: MAX_BYTES,
discard: async_nats::jetstream::stream::DiscardPolicy::Old,
max_message_size,
..Default::default()
}
}
/// Open `agent`'s terminal stream, creating it if it does not exist yet. An
/// existing stream is opened as it is.
#[cfg(feature = "agent-term")]
pub async fn open_or_create(
client: &async_nats::Client,
agent: &str,
) -> Result<async_nats::jetstream::stream::Stream, crate::Error> {
let js = async_nats::jetstream::new(client.clone());
let name = stream_name(agent);
match js.get_stream(&name).await {
Ok(stream) => Ok(stream),
Err(e) => {
tracing::info!(stream = %name, reason = %e, "agent terminal stream not available, creating it");
let max_message_size = i32::try_from(crate::max_payload(client)).unwrap_or(i32::MAX);
js.create_stream(stream_config(agent, max_message_size))
.await
.map_err(|source| crate::Error::CreateAgentStream {
stream: name,
source,
})
}
}
}
/// The consumer that replays `agent`'s rows from `start_sequence` and then
/// follows new ones, delivered to `deliver_subject` in stream order.
///
/// Ephemeral and unacknowledged: the server drops it once nothing listens
/// on `deliver_subject`, so a reader that goes away needs no delete grant.
/// `name` must be set, because the server only takes the filter from the
/// subject [`consumer_create_grant`] covers when the create names a consumer.
#[cfg(feature = "agent-term")]
#[must_use]
pub fn replay_consumer_config(
agent: &str,
name: String,
deliver_subject: String,
start_sequence: u64,
) -> async_nats::jetstream::consumer::push::Config {
async_nats::jetstream::consumer::push::Config {
name: Some(name),
deliver_subject,
filter_subject: stream_subject(agent),
deliver_policy: async_nats::jetstream::consumer::DeliverPolicy::ByStartSequence {
start_sequence,
},
ack_policy: async_nats::jetstream::consumer::AckPolicy::None,
..Default::default()
}
}
#[cfg(test)]
mod tests {
use super::*;
/// Token-wise NATS subject overlap: whether some literal subject matches
/// both patterns.
fn overlaps(a: &str, b: &str) -> bool {
let (mut a, mut b) = (a.split('.'), b.split('.'));
loop {
match (a.next(), b.next()) {
(Some(">"), Some(_)) | (Some(_), Some(">")) | (None, None) => return true,
(None, _) | (_, None) => return false,
(Some(x), Some(y)) if x != "*" && y != "*" && x != y => return false,
_ => {}
}
}
}
#[test]
fn names_match_the_publisher_and_the_grant() {
assert_eq!(stream_name("iris"), "term-main-iris");
assert_eq!(stream_subject("iris"), "$SWARM.term.iris");
assert_eq!(
consumer_create_grant(),
"$JS.API.CONSUMER.CREATE.*.*.$SWARM.term.*"
);
}
/// The server refuses to create a stream whose subjects overlap
/// another's, so this stream must share no subject with any agent's
/// subagent stream, its own agent's included.
#[test]
fn the_subject_is_disjoint_from_every_subagent_stream() {
for other in ["iris", "argus", "sub", "iris-2"] {
assert!(
!overlaps(
&stream_subject("iris"),
&crate::subagent_term::stream_subjects(other)
),
"term-main-iris overlaps term-sub-{other}"
);
}
// Control: the checker does report a real overlap.
assert!(overlaps(
"$SWARM.term.iris.>",
&crate::subagent_term::stream_subjects("iris")
));
assert!(overlaps(&stream_subject("iris"), "$SWARM.term.*"));
}
/// No agent name makes this stream's name another agent's subagent
/// stream: an agent may be called `sub-x`.
#[test]
fn the_name_never_collides_with_a_subagent_stream() {
for (agent, other) in [("sub-x", "x"), ("x", "sub-x"), ("main-x", "x")] {
assert_ne!(stream_name(agent), crate::subagent_term::stream_name(other));
}
}
/// The create the controller sends is one the grant covers, and every
/// filter the grant admits stays below `$SWARM.term`.
#[test]
fn the_grant_admits_only_filters_below_the_terminal_prefix() {
let grant = consumer_create_grant();
let create = |filter: &str| format!("$JS.API.CONSUMER.CREATE.term-main-iris.r1.{filter}");
assert!(overlaps(&grant, &create(&stream_subject("iris"))));
// The width the doc states: a literal wildcard as the last token.
assert!(overlaps(&grant, &create("$SWARM.term.>")));
for other in [
crate::subagent_term::subject("iris", "scout"),
crate::subagent_term::stream_subjects("iris"),
crate::agent_state_subject("iris"),
] {
assert!(!overlaps(&grant, &create(&other)), "{other}");
}
assert!(!overlaps(&grant, "$JS.API.CONSUMER.CREATE.term-main-iris"));
}
#[test]
fn a_fresh_reader_replays_the_newest_rows() {
assert_eq!(replay_start(1000, None), 1000 - REPLAY_ROWS + 1);
assert_eq!(replay_start(5, None), 1);
assert_eq!(replay_start(0, None), 1);
}
#[test]
fn a_reconnecting_reader_resumes_after_its_last_row() {
assert_eq!(replay_start(1000, Some(990)), 991);
assert_eq!(replay_start(1000, Some(1000)), 1001);
}
/// The consumer starts at a stored sequence, not at the next new row, so
/// the replay reaches the reader before the live tail, in stream order,
/// through the one subscription.
#[cfg(feature = "agent-term")]
#[test]
fn the_consumer_replays_before_it_follows() {
let cfg = replay_consumer_config("iris", "r1".to_owned(), "_INBOX.x".to_owned(), 901);
assert_eq!(
cfg.deliver_policy,
async_nats::jetstream::consumer::DeliverPolicy::ByStartSequence {
start_sequence: 901
}
);
assert_eq!(cfg.filter_subject, "$SWARM.term.iris");
assert_eq!(cfg.name.as_deref(), Some("r1"));
assert_eq!(
cfg.ack_policy,
async_nats::jetstream::consumer::AckPolicy::None
);
assert!(cfg.durable_name.is_none());
}
#[cfg(feature = "agent-term")]
#[test]
fn stream_config_matches_the_subagent_streams_limits() {
let cfg = stream_config("iris", 8_388_608);
assert_eq!(cfg.subjects, ["$SWARM.term.iris"]);
assert_eq!(cfg.max_age, crate::subagent_term::MAX_AGE);
assert_eq!(cfg.max_bytes, MAX_BYTES);
assert_eq!(
cfg.discard,
async_nats::jetstream::stream::DiscardPolicy::Old
);
assert_eq!(cfg.max_message_size, 8_388_608);
}
}