statserv: persist per-channel activity across restart too
All checks were successful
CI / check (push) Successful in 3m49s

The earlier StatsSet fix persisted only the shared namespaced counters. The
per-channel BOTSTATS view (lines seen + top talkers, from /statserv #chan) lived
on the LiveChannel struct, which is rebuilt from each burst — so it reset to zero
on restart. Move it to a name-keyed Network.chan_activity map (independent of live
membership, seedable before the burst) and carry it in the StatsSet snapshot
(new serde-default `channels` field) alongside the counters. Reworded the
notice from 'seen this session' now that it persists.
This commit is contained in:
Jean Chevronnet 2026-07-16 17:50:11 +00:00
parent 8a9b6a37ca
commit 1f6a11dba2
No known key found for this signature in database
8 changed files with 91 additions and 26 deletions

View file

@ -10,7 +10,7 @@ pub fn handle(me: &str, from: &Sender, chan: &str, ctx: &mut ServiceCtx, net: &d
ctx.notice(me, from.uid, format!("No activity recorded for \x02{chan}\x02 yet.")); ctx.notice(me, from.uid, format!("No activity recorded for \x02{chan}\x02 yet."));
return; return;
}; };
ctx.notice(me, from.uid, format!("\x02{chan}\x02\x02{lines}\x02 line(s) seen this session.")); ctx.notice(me, from.uid, format!("\x02{chan}\x02\x02{lines}\x02 line(s) seen."));
if top.is_empty() { if top.is_empty() {
return; return;
} }

View file

@ -136,9 +136,15 @@ pub enum Event {
// by minting a fake one on `sid`), so Local scope but still persisted. // by minting a fake one on `sid`), so Local scope but still persisted.
JupeAdded { name: String, sid: String, reason: String }, JupeAdded { name: String, sid: String, reason: String },
JupeRemoved { name: String }, JupeRemoved { name: String },
// A full snapshot of the shared stat counters (StatServ / gRPC Stats), // A full snapshot of the stats (StatServ / gRPC Stats), written periodically
// written periodically and on shutdown so they survive a restart. // and on shutdown so they survive a restart: shared namespaced counters, plus
StatsSet { counters: Vec<(String, u64)> }, // per-channel activity (channel, line count, top talkers). `channels` defaults
// for log entries written before it existed.
StatsSet {
counters: Vec<(String, u64)>,
#[serde(default)]
channels: super::ChanStats,
},
} }
// Whether an event replicates across the federation. Account identity is Global // Whether an event replicates across the federation. Account identity is Global
@ -627,9 +633,10 @@ pub(crate) fn apply(accounts: &mut HashMap<String, Account>, channels: &mut Hash
Event::JupeRemoved { name } => { Event::JupeRemoved { name } => {
net.jupes.retain(|j| !j.name.eq_ignore_ascii_case(&name)); net.jupes.retain(|j| !j.name.eq_ignore_ascii_case(&name));
} }
Event::StatsSet { counters } => { Event::StatsSet { counters, channels } => {
// A full snapshot: replace, so replaying the newest wins. // A full snapshot: replace, so replaying the newest wins.
net.stats = counters.into_iter().collect(); net.stats = counters.into_iter().collect();
net.chan_stats = channels;
} }
Event::AccountExpiryWarned { account } => { Event::AccountExpiryWarned { account } => {
if let Some(a) = accounts.get_mut(&key(&account)) { if let Some(a) = accounts.get_mut(&key(&account)) {

View file

@ -282,6 +282,9 @@ pub struct Report {
// The network-wide lists that aren't accounts/channels/grouped/bots: the network // The network-wide lists that aren't accounts/channels/grouped/bots: the network
// bans and the news items. Bundled so `apply` threads one parameter for them // bans and the news items. Bundled so `apply` threads one parameter for them
// (and stays within its argument budget as more are added). // (and stays within its argument budget as more are added).
// Persisted per-channel BOTSTATS activity: (channel, line count, top talkers).
pub type ChanStats = Vec<(String, u64, Vec<(String, u64)>)>;
#[derive(Debug, Clone, Default)] #[derive(Debug, Clone, Default)]
pub struct NetData { pub struct NetData {
pub akills: Vec<Akill>, pub akills: Vec<Akill>,
@ -310,6 +313,8 @@ pub struct NetData {
// Persisted shared stat counters (StatServ / gRPC Stats), snapshotted so they // Persisted shared stat counters (StatServ / gRPC Stats), snapshotted so they
// survive a restart (live gauges are re-derived, not stored here). // survive a restart (live gauges are re-derived, not stored here).
pub stats: std::collections::BTreeMap<String, u64>, pub stats: std::collections::BTreeMap<String, u64>,
// Persisted per-channel BOTSTATS activity: (channel, lines, top talkers).
pub chan_stats: ChanStats,
} }
// A session-limit exception: an IP-mask glob and the session allowance for IPs // A session-limit exception: an IP-mask glob and the session allowance for IPs
@ -1168,8 +1173,11 @@ impl Db {
if self.net.default_bot.is_some() { if self.net.default_bot.is_some() {
snapshot.push(Event::DefaultBotSet { bot: self.net.default_bot.clone() }); snapshot.push(Event::DefaultBotSet { bot: self.net.default_bot.clone() });
} }
if !self.net.stats.is_empty() { if !self.net.stats.is_empty() || !self.net.chan_stats.is_empty() {
snapshot.push(Event::StatsSet { counters: self.net.stats.iter().map(|(k, v)| (k.clone(), *v)).collect() }); snapshot.push(Event::StatsSet {
counters: self.net.stats.iter().map(|(k, v)| (k.clone(), *v)).collect(),
channels: self.net.chan_stats.clone(),
});
} }
for host in &self.host_cfg.offers { for host in &self.host_cfg.offers {
snapshot.push(Event::VhostOfferAdded { host: host.clone() }); snapshot.push(Event::VhostOfferAdded { host: host.clone() });

View file

@ -250,10 +250,24 @@ impl Db {
self.net.stats.clone() self.net.stats.clone()
} }
/// Snapshot the live stat counters to the log so they survive a restart. /// The persisted per-channel activity, to seed BOTSTATS on startup.
pub fn persist_stats(&mut self, counters: &std::collections::BTreeMap<String, u64>) -> std::io::Result<()> { pub fn persisted_chan_stats(&self) -> ChanStats {
self.log.append(Event::StatsSet { counters: counters.iter().map(|(k, v)| (k.clone(), *v)).collect() })?; self.net.chan_stats.clone()
}
/// Snapshot the live stats (shared counters + per-channel activity) to the log
/// so they survive a restart.
pub fn persist_stats(
&mut self,
counters: &std::collections::BTreeMap<String, u64>,
chan_stats: ChanStats,
) -> std::io::Result<()> {
self.log.append(Event::StatsSet {
counters: counters.iter().map(|(k, v)| (k.clone(), *v)).collect(),
channels: chan_stats.clone(),
})?;
self.net.stats = counters.clone(); self.net.stats = counters.clone();
self.net.chan_stats = chan_stats;
Ok(()) Ok(())
} }

View file

@ -532,13 +532,15 @@
let p = tmp("statspersist"); let p = tmp("statspersist");
let counters: std::collections::BTreeMap<String, u64> = let counters: std::collections::BTreeMap<String, u64> =
[("chanserv.register".to_string(), 7u64), ("nickserv.identify".to_string(), 42u64)].into_iter().collect(); [("chanserv.register".to_string(), 7u64), ("nickserv.identify".to_string(), 42u64)].into_iter().collect();
let chans = vec![("#devs".to_string(), 6u64, vec![("reverse".to_string(), 5u64), ("Keiko".to_string(), 1u64)])];
{ {
let mut db = Db::open(&p, "N1"); let mut db = Db::open(&p, "N1");
db.persist_stats(&counters).unwrap(); db.persist_stats(&counters, chans.clone()).unwrap();
assert_eq!(db.persisted_stats(), counters, "stored while live"); assert_eq!(db.persisted_stats(), counters, "stored while live");
} }
let db = Db::open(&p, "N1"); let db = Db::open(&p, "N1");
assert_eq!(db.persisted_stats(), counters, "counters replay from the log after a restart"); assert_eq!(db.persisted_stats(), counters, "counters replay from the log after a restart");
assert_eq!(db.persisted_chan_stats(), chans, "per-channel activity replays too");
} }
// Fold parity: the live state after a broad sequence of writes must equal the // Fold parity: the live state after a broad sequence of writes must equal the

View file

@ -213,8 +213,9 @@ impl Engine {
(!topics.is_empty()).then(|| (s.nick().to_string(), blurb, topics)) (!topics.is_empty()).then(|| (s.nick().to_string(), blurb, topics))
}) })
.collect(); .collect();
// Restore the persisted stat counters so they survive a restart. // Restore persisted stats so they survive a restart.
network.seed_stats(db.persisted_stats()); network.seed_stats(db.persisted_stats());
network.seed_chan_activity(db.persisted_chan_stats());
Self { Self {
services, services,
network, network,
@ -275,10 +276,11 @@ impl Engine {
// Only the real counters are persisted; the live gauges above are re-derived. // Only the real counters are persisted; the live gauges above are re-derived.
pub fn persist_stats(&mut self) { pub fn persist_stats(&mut self) {
let counters = self.network.stat_counters().clone(); let counters = self.network.stat_counters().clone();
if counters.is_empty() { let chan_stats = self.network.chan_activity_snapshot();
if counters.is_empty() && chan_stats.is_empty() {
return; return;
} }
if let Err(err) = self.db.persist_stats(&counters) { if let Err(err) = self.db.persist_stats(&counters, chan_stats) {
tracing::warn!(%err, "failed to persist stat counters"); tracing::warn!(%err, "failed to persist stat counters");
} }
} }

View file

@ -23,9 +23,12 @@ pub struct Network {
// Downstream server tree: SID -> the SID that introduced it. Lets a hub's // Downstream server tree: SID -> the SID that introduced it. Lets a hub's
// SQUIT cascade to every server (and user) behind it. Rebuilt each link. // SQUIT cascade to every server (and user) behind it. Rebuilt each link.
servers: HashMap<String, String>, servers: HashMap<String, String>,
// Shared, namespaced stat counters any service contributes to (ephemeral, // Shared, namespaced stat counters any service contributes to (persisted via
// ordered for a stable snapshot). Read by StatServ and the gRPC Stats API. // StatsSet). Read by StatServ and the gRPC Stats API.
stats: BTreeMap<String, u64>, stats: BTreeMap<String, u64>,
// Per-channel BOTSTATS activity (lines + top talkers), name-keyed and
// persisted alongside `stats` so it survives a restart.
chan_activity: HashMap<String, ChanActivity>,
// Live session count per connecting IP, for OperServ session limiting. // Live session count per connecting IP, for OperServ session limiting.
sessions: HashMap<String, u32>, sessions: HashMap<String, u32>,
// Recent moderation/action incidents (bounded ring), for OperServ LOGSEARCH. // Recent moderation/action incidents (bounded ring), for OperServ LOGSEARCH.
@ -85,7 +88,13 @@ pub struct Channel {
pub ops: HashSet<String>, // uids holding channel-operator status pub ops: HashSet<String>, // uids holding channel-operator status
pub voices: HashSet<String>, // uids holding +v pub voices: HashSet<String>, // uids holding +v
pub key: Option<String>, pub key: Option<String>,
// BOTSTATS: lines seen this session, and per-nick counts (top-talkers). Bounded. }
// Per-channel BOTSTATS activity: total lines and per-nick counts (top talkers).
// Held name-keyed (not on the live Channel, which is rebuilt from each burst) and
// persisted, so a channel's history survives a services restart.
#[derive(Default, Clone)]
pub struct ChanActivity {
pub lines: u64, pub lines: u64,
pub talkers: HashMap<String, u64>, pub talkers: HashMap<String, u64>,
} }
@ -332,10 +341,10 @@ impl Network {
// is capped so a busy channel can't grow it without bound. // is capped so a busy channel can't grow it without bound.
pub fn record_line(&mut self, channel: &str, nick: &str) { pub fn record_line(&mut self, channel: &str, nick: &str) {
const TALKER_CAP: usize = 512; const TALKER_CAP: usize = 512;
let c = self.channels.entry(lc(channel)).or_default(); let a = self.chan_activity.entry(lc(channel)).or_default();
c.lines = c.lines.saturating_add(1); a.lines = a.lines.saturating_add(1);
if c.talkers.len() < TALKER_CAP || c.talkers.contains_key(nick) { if a.talkers.len() < TALKER_CAP || a.talkers.contains_key(nick) {
*c.talkers.entry(nick.to_string()).or_insert(0) += 1; *a.talkers.entry(nick.to_string()).or_insert(0) += 1;
} }
} }
@ -356,11 +365,34 @@ impl Network {
// BOTSTATS view: (total lines, top talkers by count, descending). // BOTSTATS view: (total lines, top talkers by count, descending).
pub fn channel_activity(&self, channel: &str) -> Option<(u64, Vec<(String, u64)>)> { pub fn channel_activity(&self, channel: &str) -> Option<(u64, Vec<(String, u64)>)> {
let c = self.channels.get(&lc(channel))?; let a = self.chan_activity.get(&lc(channel))?;
let mut top: Vec<(String, u64)> = c.talkers.iter().map(|(n, v)| (n.clone(), *v)).collect(); let mut top: Vec<(String, u64)> = a.talkers.iter().map(|(n, v)| (n.clone(), *v)).collect();
top.sort_by(|a, b| b.1.cmp(&a.1).then_with(|| a.0.cmp(&b.0))); top.sort_by(|x, y| y.1.cmp(&x.1).then_with(|| x.0.cmp(&y.0)));
top.truncate(10); top.truncate(10);
Some((c.lines, top)) Some((a.lines, top))
}
// Seed per-channel activity from persisted state at startup.
pub fn seed_chan_activity(&mut self, data: crate::engine::db::ChanStats) {
self.chan_activity = data
.into_iter()
.map(|(c, lines, talkers)| (c, ChanActivity { lines, talkers: talkers.into_iter().collect() }))
.collect();
}
// Snapshot per-channel activity for persistence: each active channel's line
// count and its top talkers (capped, so a busy channel bounds the event).
pub fn chan_activity_snapshot(&self) -> crate::engine::db::ChanStats {
self.chan_activity
.iter()
.filter(|(_, a)| a.lines > 0)
.map(|(c, a)| {
let mut talkers: Vec<(String, u64)> = a.talkers.iter().map(|(n, v)| (n.clone(), *v)).collect();
talkers.sort_by(|x, y| y.1.cmp(&x.1).then_with(|| x.0.cmp(&y.0)));
talkers.truncate(50);
(c.clone(), a.lines, talkers)
})
.collect()
} }
// Uids currently in `channel`. // Uids currently in `channel`.

View file

@ -2723,7 +2723,7 @@
// Per-channel view. // Per-channel view.
let out = ss(&mut e, "#c"); let out = ss(&mut e, "#c");
assert!(out.iter().any(|a| matches!(a, NetAction::Notice { text, .. } if text.contains("line(s) seen this session"))), "channel lines: {out:?}"); assert!(out.iter().any(|a| matches!(a, NetAction::Notice { text, .. } if text.contains("line(s) seen"))), "channel lines: {out:?}");
assert!(out.iter().any(|a| matches!(a, NetAction::Notice { text, .. } if text.contains("spammer") && text.contains("3"))), "top talker: {out:?}"); assert!(out.iter().any(|a| matches!(a, NetAction::Notice { text, .. } if text.contains("spammer") && text.contains("3"))), "top talker: {out:?}");
// Global view (oper) shows the shared counters. // Global view (oper) shows the shared counters.
let out = ss(&mut e, "SERVER"); let out = ss(&mut e, "SERVER");