MF-11 update-return (0.3.21), the app half on release-0.3.21: the Windows update helper owns the return (the exe set kept beside the app, the app launched by the helper, api/state polled 120 s, the kept set restored when nothing answers, one intake line either way; the installer stays silent under /IGNOTA=2), the window host restarts a dead engine and answers the Restart Manager, the first act after an update is the read-back line and a FAULT pc-restart line when the PC came up from a power loss; the tuner never asks above a card's measured efficient point (the 5090 at 308 W, floored at the vendor's 400 W) unless Power control is on and the user raised the cap, and a refused cap is asked again at 2 and 10 min then reported as FAULT power-cap; the job-channel ping on every wake request (7 October 2026, PC 2 lost power at 10:46Z; the relay half stays on update-return)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs
parent
db4212ca52
commit
e68ba08ccc
9 changed files with 752 additions and 57 deletions
73
app/igneum-app/src/bootcheck.rs
Normal file
73
app/igneum-app/src/bootcheck.rs
Normal file
|
|
@ -0,0 +1,73 @@
|
|||
//! Did this PC come up from a power loss or a hard reset? (MF-11, 7 October 2026: PC 2 dropped twice in one day with
|
||||
//! Kernel-Power 41 and EventLog 6008 at the next boot, no bugcheck, no dump, and nothing said so until a person read
|
||||
//! the event log.) Windows: the System log's event 41 (Kernel-Power, critical) or 6008 (EventLog, "the previous
|
||||
//! shutdown was unexpected") inside the last 15 minutes, read once at the engine's start through wevtutil; the engine
|
||||
//! logs one `FAULT pc-restart:` line to the intake. Other platforms: nothing (a Mac's power log is not this class).
|
||||
|
||||
use std::process::Command;
|
||||
use std::time::Duration;
|
||||
|
||||
/// How far back the start-up check looks: an engine starts at login, inside a minute or two of the boot.
|
||||
pub const WINDOW_MS: u64 = 15 * 60 * 1000;
|
||||
|
||||
/// One line naming the event, or None when the boot was clean, the query failed, or this is not Windows.
|
||||
pub fn unexpected_restart() -> Option<String> {
|
||||
if !cfg!(windows) {
|
||||
return None;
|
||||
}
|
||||
let query = format!("*[System[(EventID=41 or EventID=6008) and TimeCreated[timediff(@SystemTime) <= {WINDOW_MS}]]]");
|
||||
let mut c = Command::new(crate::platform::tool("wevtutil"));
|
||||
c.args(["qe", "System", &format!("/q:{query}"), "/f:text", "/c:2", "/rd:true"]);
|
||||
let out = crate::detect::run_timeout(&mut c, None, Duration::from_secs(20))?;
|
||||
parse_events(&out)
|
||||
}
|
||||
|
||||
/// The reading of wevtutil's text output: the newest 41 or 6008 as "event 41 (Kernel-Power) at <time>" or
|
||||
/// "event 6008 (the previous shutdown was unexpected) at <time>"; None when neither is in the text.
|
||||
pub fn parse_events(text: &str) -> Option<String> {
|
||||
let mut date = String::new();
|
||||
let mut id = String::new();
|
||||
for line in text.lines() {
|
||||
let t = line.trim();
|
||||
if let Some(d) = t.strip_prefix("Date:") {
|
||||
date = d.trim().to_string();
|
||||
} else if let Some(i) = t.strip_prefix("Event ID:") {
|
||||
id = i.trim().to_string();
|
||||
if id == "41" || id == "6008" {
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
match id.as_str() {
|
||||
"41" => Some(format!("event 41 (Kernel-Power: the system rebooted without shutting down cleanly) at {date}")),
|
||||
"6008" => Some(format!("event 6008 (the previous shutdown was unexpected) at {date}")),
|
||||
_ => None,
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::parse_events;
|
||||
|
||||
const PC2: &str = "Event[0]\n Log Name: System\n Source: Microsoft-Windows-Kernel-Power\n Date: 2026-10-07T14:37:19.4360000Z\n Event ID: 41\n Task: N/A\n Level: Critical\n Opcode: Info\n Keyword: N/A\n User: S-1-5-18\n Computer: X\n Description: \n\nEvent[1]\n Log Name: System\n Source: EventLog\n Date: 2026-10-07T14:37:29.0380000Z\n Event ID: 6008\n Task: None\n Level: Error\n";
|
||||
|
||||
/// PC 2's boot of 13:37Z on 7 October 2026 (the collection job's wevtutil text): the known-failed case reads as one.
|
||||
#[test]
|
||||
fn pc2_boot_reads_as_a_power_loss() {
|
||||
assert_eq!(parse_events(PC2), Some("event 41 (Kernel-Power: the system rebooted without shutting down cleanly) at 2026-10-07T14:37:19.4360000Z".into()));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn a_6008_alone_is_the_unexpected_shutdown() {
|
||||
let t = "Event[0]\n Log Name: System\n Source: EventLog\n Date: 2026-10-07T08:22:34.0000000Z\n Event ID: 6008\n Level: Error\n";
|
||||
assert_eq!(parse_events(t), Some("event 6008 (the previous shutdown was unexpected) at 2026-10-07T08:22:34.0000000Z".into()));
|
||||
}
|
||||
|
||||
/// A clean boot (the known-good case): nothing, including other ids inside the window and an empty answer.
|
||||
#[test]
|
||||
fn a_clean_boot_reads_as_nothing() {
|
||||
assert_eq!(parse_events(""), None);
|
||||
assert_eq!(parse_events("Event[0]\n Log Name: System\n Source: Microsoft-Windows-Kernel-General\n Date: 2026-10-07T14:37:17.7400000Z\n Event ID: 12\n Level: Information\n"), None);
|
||||
assert_eq!(parse_events("wevtutil: access denied"), None);
|
||||
}
|
||||
}
|
||||
|
|
@ -25,6 +25,31 @@ pub const POWER_STEPS_PCT: [u32; 6] = [100, 90, 80, 70, 60, 50];
|
|||
pub const CLOCK_STEPS_PCT: [u32; 7] = [100, 90, 80, 70, 60, 50, 45];
|
||||
/// A card's clock floor when the vendor reports none: this share of its maximum core clock.
|
||||
pub const CLOCK_FLOOR_PCT: u32 = 45;
|
||||
/// The measured efficient point per card (docs/bench-log.md: the 5090 at 308 W for 122 MH/s, the 9070 XT at 199 W for
|
||||
/// 18 MH/s), the tuner's ceiling. Rule (main, 7 October 2026, after PC 2 dropped nine minutes after the tuner asked its
|
||||
/// 5090 for 575 W with proving on the same card): the tuner never requests more than the card's measured efficient point
|
||||
/// unless Power control is on and the user raised the cap above the default; a card not in this table is held at the
|
||||
/// cap it runs at. A point below the vendor's minimum limit clamps up to that minimum (the 5090 floors at 400 W).
|
||||
pub const EFFICIENT_W: &[(&str, f64)] = &[("RTX 5090", 308.0), ("RX 9070 XT", 199.0)];
|
||||
/// The app's cap when the user chose none (engine.rs requested_watts); a cap above it is one the user raised.
|
||||
pub const DEFAULT_CAP_PCT: u32 = 80;
|
||||
|
||||
/// The measured efficient watts of a card by its name, when the table has it.
|
||||
pub fn efficient_watts(name: &str) -> Option<f64> {
|
||||
let n = name.to_ascii_uppercase();
|
||||
EFFICIENT_W.iter().find(|(k, _)| n.contains(&k.to_ascii_uppercase())).map(|(_, w)| *w)
|
||||
}
|
||||
|
||||
/// The tuner's power ceiling in watts for a card: its measured efficient point, or the cap it runs at when the table has
|
||||
/// no entry; the user's own cap instead when Power control is on and that cap is above the default (they raised it).
|
||||
pub fn power_ceiling(limits: &Limits, name: &str, before: Point, power_control: bool) -> f64 {
|
||||
let cap = Limits { power_ceiling_w: 0.0, ..limits.clone() }.watts_for(if before.power_pct == 0 { DEFAULT_CAP_PCT } else { before.power_pct });
|
||||
let raised = power_control && before.power_pct > DEFAULT_CAP_PCT;
|
||||
match efficient_watts(name) {
|
||||
Some(e) if !raised => e.min(cap),
|
||||
_ => cap,
|
||||
}
|
||||
}
|
||||
/// A point may lose this much rate against the fastest point and still win on MH per watt (the manifest can change it).
|
||||
pub const RATE_TOLERANCE_PCT: f64 = 1.0;
|
||||
/// A step whose hottest GPU reading reaches this is marked hot and cannot win (the engine aborts at 90).
|
||||
|
|
@ -55,9 +80,20 @@ pub struct Limits {
|
|||
/// clocks.max.mem); 0 = no memory knob (AMD through ADLX on RDNA 4 exposes none)
|
||||
pub mem_default_mhz: u32,
|
||||
pub mem_max_mhz: u32,
|
||||
/// the tuner's ceiling (power_ceiling), 0 = none: no step asks for more watts than this
|
||||
pub power_ceiling_w: f64,
|
||||
}
|
||||
|
||||
impl Limits {
|
||||
/// The ceiling as the vendor allows it: never below the card's minimum limit; 0 when there is none.
|
||||
pub fn ceiling(&self) -> f64 {
|
||||
if self.power_ceiling_w <= 0.0 { 0.0 } else { self.power_ceiling_w.max(self.power_min_w) }
|
||||
}
|
||||
/// The percent of the default that the ceiling is, for the first step of the power ladder.
|
||||
pub fn ceiling_pct(&self) -> u32 {
|
||||
let c = self.ceiling();
|
||||
if c <= 0.0 || self.power_default_w <= 0.0 { 100 } else { (c / self.power_default_w * 100.0).round().clamp(1.0, 100.0) as u32 }
|
||||
}
|
||||
pub fn clock_floor(&self) -> u32 {
|
||||
if self.clock_min_mhz > 0 {
|
||||
self.clock_min_mhz
|
||||
|
|
@ -88,6 +124,10 @@ impl Limits {
|
|||
if self.power_max_w > 0.0 {
|
||||
w = w.min(self.power_max_w);
|
||||
}
|
||||
let c = self.ceiling();
|
||||
if c > 0.0 {
|
||||
w = w.min(c);
|
||||
}
|
||||
w.round()
|
||||
}
|
||||
}
|
||||
|
|
@ -276,7 +316,9 @@ impl Plan {
|
|||
pub fn full(limits: &Limits, before: Point, tolerance_pct: f64) -> Plan {
|
||||
let mut power = Vec::new();
|
||||
if limits.power_default_w > 0.0 {
|
||||
for pct in POWER_STEPS_PCT {
|
||||
// the ceiling first (its own percent), then the ladder; every step above it clamps to it and is dropped as a duplicate
|
||||
let top = if limits.ceiling() > 0.0 { vec![limits.ceiling_pct()] } else { Vec::new() };
|
||||
for pct in top.into_iter().chain(POWER_STEPS_PCT) {
|
||||
let w = limits.watts_for(pct);
|
||||
if power.last().map(|s: &Step| (s.watts - w).abs() < 0.5).unwrap_or(false) {
|
||||
continue;
|
||||
|
|
@ -1051,7 +1093,71 @@ mod tests {
|
|||
/// PC 1's RTX 5090 (nvidia-smi, 4 and 5 October 2026): default 575 W, min 400 W, max 600 W; clocks.max.gr is
|
||||
/// read at the first tune (3,090 MHz is the shape used here, not a measurement).
|
||||
fn l5090() -> Limits {
|
||||
Limits { power_default_w: 575.0, power_min_w: 400.0, power_max_w: 600.0, clock_max_mhz: 3090, clock_min_mhz: 0, mem_default_mhz: 13801, mem_max_mhz: 14001 }
|
||||
Limits { power_default_w: 575.0, power_min_w: 400.0, power_max_w: 600.0, clock_max_mhz: 3090, clock_min_mhz: 0, mem_default_mhz: 13801, mem_max_mhz: 14001, power_ceiling_w: 0.0 }
|
||||
}
|
||||
|
||||
fn with_ceiling(name: &str, before: Point, power_control: bool) -> Limits {
|
||||
let mut l = l5090();
|
||||
l.power_ceiling_w = power_ceiling(&l, name, before, power_control);
|
||||
l
|
||||
}
|
||||
fn max_watts(plan: &Plan) -> f64 {
|
||||
let mut rows = Vec::new();
|
||||
let mut top: f64 = 0.0;
|
||||
while let Some(s) = plan.next(&rows) {
|
||||
top = top.max(s.watts);
|
||||
rows.push(row_at(s.point, s.watts, 100.0));
|
||||
if rows.len() > 40 { break; }
|
||||
}
|
||||
top
|
||||
}
|
||||
|
||||
/// PC 2, 7 October 2026: the tuner asked the 5090 for 100% (575 W) while proving ran on the same card; the PC dropped
|
||||
/// nine minutes later. The known-failed shape first: without a ceiling the full plan opens at 575 W.
|
||||
#[test]
|
||||
fn without_a_ceiling_the_full_plan_asks_the_5090_for_575_w() {
|
||||
let plan = Plan::full(&l5090(), Point { clock_mhz: 0, power_pct: 70, mem_mhz: 0 }, 1.0);
|
||||
assert_eq!(plan.next(&[]).unwrap().watts, 575.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn the_tuner_never_asks_above_the_measured_efficient_point() {
|
||||
// the 5090's measured point is 308 W; the vendor's minimum is 400 W, so the ceiling clamps up to 400
|
||||
let l = with_ceiling("NVIDIA GeForce RTX 5090", Point { clock_mhz: 0, power_pct: 70, mem_mhz: 0 }, false);
|
||||
assert_eq!(l.power_ceiling_w, 308.0);
|
||||
assert_eq!(l.ceiling(), 400.0);
|
||||
let plan = Plan::full(&l, Point { clock_mhz: 0, power_pct: 70, mem_mhz: 0 }, 1.0);
|
||||
let first = plan.next(&[]).unwrap();
|
||||
assert_eq!((first.watts, first.point.power_pct, first.kind), (400.0, 70, Kind::Power), "one power step at the ceiling, with its own percent");
|
||||
assert_eq!(plan.len(), 1 + 6, "one power step (every ladder step clamps to 400 W) and the six clock steps");
|
||||
assert!(max_watts(&plan) <= 400.0, "no step above the ceiling");
|
||||
// the confirm plan and the climb honour it too: a fleet prior at 100% is clamped
|
||||
let prior = Point { clock_mhz: 0, power_pct: 100, mem_mhz: 0 };
|
||||
assert!(max_watts(&Plan::confirm(&l, prior, Point { clock_mhz: 0, power_pct: 70, mem_mhz: 0 }, 1.0)) <= 400.0);
|
||||
assert!(max_watts(&Plan::climb(&l, prior, Goal::Efficiency, 1.0)) <= 400.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn power_control_on_with_a_cap_the_user_raised_lifts_the_ceiling_to_that_cap() {
|
||||
let l = with_ceiling("NVIDIA GeForce RTX 5090", Point { clock_mhz: 0, power_pct: 90, mem_mhz: 0 }, true);
|
||||
assert_eq!(l.ceiling(), 518.0, "the user's 90% (518 W), never the vendor's 575 W");
|
||||
assert!(max_watts(&Plan::full(&l, Point { clock_mhz: 0, power_pct: 90, mem_mhz: 0 }, 1.0)) <= 518.0);
|
||||
// Power control on at the default cap is not a raise: the measured point holds
|
||||
let d = with_ceiling("NVIDIA GeForce RTX 5090", Point { clock_mhz: 0, power_pct: DEFAULT_CAP_PCT, mem_mhz: 0 }, true);
|
||||
assert_eq!(d.ceiling(), 400.0);
|
||||
// a raised cap with Power control OFF does not count
|
||||
let off = with_ceiling("NVIDIA GeForce RTX 5090", Point { clock_mhz: 0, power_pct: 90, mem_mhz: 0 }, false);
|
||||
assert_eq!(off.ceiling(), 400.0);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn a_card_not_in_the_table_is_held_at_the_cap_it_runs_at() {
|
||||
let l = with_ceiling("NVIDIA GeForce RTX 4070", Point { clock_mhz: 0, power_pct: 70, mem_mhz: 0 }, false);
|
||||
assert_eq!(l.ceiling(), 403.0, "70% of 575 W, the card's own cap");
|
||||
let zero = with_ceiling("NVIDIA GeForce RTX 4070", Point::default(), false);
|
||||
assert_eq!(zero.ceiling(), 460.0, "no chosen cap: the app's default 80%");
|
||||
assert_eq!(efficient_watts("AMD Radeon RX 9070 XT"), Some(199.0));
|
||||
assert_eq!(efficient_watts("gfx1036"), None);
|
||||
}
|
||||
|
||||
fn row_at(p: Point, watts: f64, mhs: f64) -> Row {
|
||||
|
|
@ -1258,7 +1364,7 @@ mod tests {
|
|||
fn a_fault_during_a_step_reverts_it_and_the_run_goes_on() {
|
||||
let t0 = Instant::now();
|
||||
let timing = Timing { settle: Duration::from_secs(2), hold: Duration::from_secs(4), apply: Duration::from_secs(30) };
|
||||
let limits = Limits { power_default_w: 300.0, power_min_w: 150.0, power_max_w: 300.0, clock_max_mhz: 0, clock_min_mhz: 0, mem_default_mhz: 0, mem_max_mhz: 0 };
|
||||
let limits = Limits { power_default_w: 300.0, power_min_w: 150.0, power_max_w: 300.0, clock_max_mhz: 0, clock_min_mhz: 0, mem_default_mhz: 0, mem_max_mhz: 0, power_ceiling_w: 0.0 };
|
||||
let plan = Plan::full(&limits, Point { clock_mhz: 0, power_pct: 80, mem_mhz: 0 }, 1.0);
|
||||
let mut run = Run::new(0, "0", "c", plan, 240.0, false, timing, t0);
|
||||
let mut t = t0;
|
||||
|
|
|
|||
|
|
@ -719,6 +719,9 @@ pub struct Engine {
|
|||
amd_telemetry_retry_at: Instant,
|
||||
power_busy: bool,
|
||||
power_restore_pending: bool,
|
||||
/// a cap the card did not take: how many times in a row, and when to ask again (cap_refusal_plan)
|
||||
cap_refusals: u32,
|
||||
cap_retry_at: Option<Instant>,
|
||||
/// an elevated step handed to the window host: (command line, what, requested watts per device, since)
|
||||
power_via_host: Option<(String, String, std::collections::HashMap<String, f64>, Instant)>,
|
||||
/// the manifest's consensus override changed: restart the node at a safe moment
|
||||
|
|
@ -880,6 +883,8 @@ impl Engine {
|
|||
amd_telemetry_retry_at: now,
|
||||
power_busy: false,
|
||||
power_restore_pending: false,
|
||||
cap_refusals: 0,
|
||||
cap_retry_at: None,
|
||||
power_via_host: None,
|
||||
node_override_restart: false,
|
||||
node_override_unusable: false,
|
||||
|
|
@ -960,6 +965,16 @@ impl Engine {
|
|||
let v = crate::detect::node_version(&self.bins.node);
|
||||
self.st().node.version = v.clone();
|
||||
self.shared.log(&self.shared.upload_header());
|
||||
// MF-11 (7 October 2026): the first act after an update is the read-back line, so the intake shows the new
|
||||
// version up before anything else happens; the shipper's rollout table reads it per machine
|
||||
if let Some(from) = self.ota.updated_from() {
|
||||
self.shared.log(&format!("update-return: app {VERSION} up after the update from {from} (node {v}, machine {})", self.shared.runtime.id8()));
|
||||
}
|
||||
// a PC that lost power or hard-reset is a fault row at its next start, not a silence (PC 2, 7 October 2026)
|
||||
if let Some(line) = crate::bootcheck::unexpected_restart() {
|
||||
self.shared.log(&format!("FAULT pc-restart: {line}"));
|
||||
self.shared.event("error", &format!("This PC restarted without a shutdown: {line}. Mining resumed; if it happens again check the power supply and the wall."));
|
||||
}
|
||||
// which intake and which downloads folder this build reports to and checks (fingerprints, never the values;
|
||||
// rotation phase 2 reads this line from every machine's upload: docs/plans/rotation-phase-2.md)
|
||||
self.shared.log(&self.shared.packaged.describe());
|
||||
|
|
@ -1289,6 +1304,22 @@ impl Engine {
|
|||
Err(e) => e.clone(),
|
||||
};
|
||||
self.shared.event("error", &format!("GPU power cap NOT applied ({why}): {}. Retry from the card tile.", missing.join("; ")));
|
||||
// never left as "cap NOT applied" (main's rule, 7 October 2026): asked again on the ladder, then a FAULT line home
|
||||
self.cap_refusals += 1;
|
||||
match cap_refusal_plan(self.cap_refusals) {
|
||||
Some(wait) => {
|
||||
self.cap_retry_at = Some(Instant::now() + wait);
|
||||
self.shared.log(&format!("power cap: refusal {} ({why}); asking again in {} s", self.cap_refusals, wait.as_secs()));
|
||||
}
|
||||
None => {
|
||||
self.cap_retry_at = None;
|
||||
let power_control = self.shared.settings.lock().unwrap().power_control;
|
||||
self.shared.log(&format!("FAULT power-cap: {} ({why}); refused {} times, the card runs at the limit it reports; Power control is {}", missing.join("; "), self.cap_refusals, if power_control { "on" } else { "off" }));
|
||||
}
|
||||
}
|
||||
} else if !applied.is_empty() {
|
||||
self.cap_refusals = 0;
|
||||
self.cap_retry_at = None;
|
||||
}
|
||||
if applied.is_empty() && missing.is_empty() {
|
||||
self.shared.log(&format!("power cap: nothing to read back for {what}"));
|
||||
|
|
@ -1578,6 +1609,14 @@ impl Engine {
|
|||
self.shared.log(&format!("tune: request {seq} refused: {}", short(&text.replace('\n', " "), 300)));
|
||||
if self.sweep.as_ref().map(|r| r.seq == seq).unwrap_or(false) {
|
||||
self.sweep_abort("the card refused the setting (see the log)");
|
||||
} else if self.sweep.is_none() {
|
||||
// the restore after a stopped tune was refused (PC 2, 7 October 2026: request 0 back to 70% never ran):
|
||||
// the card may sit at the vendor default; the cap path asks again and reports a FAULT on its ladder
|
||||
self.shared.log(&format!("FAULT power-cap: the restore after the tune was refused ({}); re-applying the cap", short(&text.replace('\n', " "), 120)));
|
||||
for c in self.st().mining.cards.iter_mut().filter(|c| c.vendor == "nvidia") {
|
||||
c.power_applied = false;
|
||||
}
|
||||
self.apply_power_limits("tune restore refused");
|
||||
}
|
||||
}
|
||||
},
|
||||
|
|
@ -2982,11 +3021,19 @@ impl Engine {
|
|||
let power_control = probe.direct || self.shared.settings.lock().unwrap().power_control || (cfg!(windows) && self.power_task_registered());
|
||||
// AMD's power limit is a percent offset from the default (ADLX): the plan's watts scale becomes a percent
|
||||
// scale (default 100, floor 100 + plimit_min, ceiling 100 + plimit_max), tune_apply sends pct - 100
|
||||
let limits = if c.vendor == "amd" && probe.direct && probe.plimit_max >= probe.plimit_min && probe.plimit_min > -100.0 {
|
||||
crate::ember::Limits { power_default_w: 100.0, power_min_w: 100.0 + probe.plimit_min, power_max_w: 100.0 + probe.plimit_max, clock_max_mhz: probe.clock_max_mhz, clock_min_mhz: probe.clock_min_mhz, mem_default_mhz: 0, mem_max_mhz: 0 }
|
||||
let mut limits = if c.vendor == "amd" && probe.direct && probe.plimit_max >= probe.plimit_min && probe.plimit_min > -100.0 {
|
||||
crate::ember::Limits { power_default_w: 100.0, power_min_w: 100.0 + probe.plimit_min, power_max_w: 100.0 + probe.plimit_max, clock_max_mhz: probe.clock_max_mhz, clock_min_mhz: probe.clock_min_mhz, mem_default_mhz: 0, mem_max_mhz: 0, power_ceiling_w: 0.0 }
|
||||
} else {
|
||||
crate::ember::Limits { power_default_w: c.power_default_w, power_min_w: c.power_min_w, power_max_w: c.power_max_w, clock_max_mhz: probe.clock_max_mhz, clock_min_mhz: probe.clock_min_mhz, mem_default_mhz: probe.mem_default_mhz, mem_max_mhz: probe.mem_max_mhz }
|
||||
crate::ember::Limits { power_default_w: c.power_default_w, power_min_w: c.power_min_w, power_max_w: c.power_max_w, clock_max_mhz: probe.clock_max_mhz, clock_min_mhz: probe.clock_min_mhz, mem_default_mhz: probe.mem_default_mhz, mem_max_mhz: probe.mem_max_mhz, power_ceiling_w: 0.0 }
|
||||
};
|
||||
// the tuner's ceiling (main's rule, 7 October 2026, PC 2): the card's measured efficient point, or the cap it runs
|
||||
// at, unless Power control is on in Settings and the user raised the cap above the default; watts only (NVIDIA)
|
||||
if c.vendor != "amd" || !probe.direct {
|
||||
let settings_power_control = self.shared.settings.lock().unwrap().power_control;
|
||||
let at = crate::ember::Point { clock_mhz: c.clock_cap_mhz, power_pct: c.power_pct, mem_mhz: c.mem_cap_mhz };
|
||||
limits.power_ceiling_w = crate::ember::power_ceiling(&limits, &c.name, at, settings_power_control);
|
||||
self.shared.log(&format!("tune: ceiling {:.0} W for {} ({}; Power control {})", limits.ceiling(), c.name, crate::ember::efficient_watts(&c.name).map(|w| format!("measured efficient point {w:.0} W")).unwrap_or_else(|| "no measured point, the cap it runs at".into()), if settings_power_control { "on" } else { "off" }));
|
||||
}
|
||||
let control = match c.vendor.as_str() {
|
||||
"nvidia" => crate::ember::control_reason("nvidia", &limits, &c.device, power_control, false),
|
||||
"amd" => crate::ember::control_reason("amd", &limits, &c.device, power_control, probe.direct && probe.amd_ordinal >= 0),
|
||||
|
|
@ -3595,6 +3642,13 @@ impl Engine {
|
|||
self.last_awake = now;
|
||||
crate::platform::keep_awake_tick();
|
||||
}
|
||||
if self.cap_retry_at.map(|t| now >= t).unwrap_or(false) && !self.power_busy && self.sweep.is_none() && self.sweep_pending.is_none() {
|
||||
self.cap_retry_at = None;
|
||||
for c in self.st().mining.cards.iter_mut().filter(|c| c.vendor == "nvidia" && !c.power_applied) {
|
||||
c.power_applied = false;
|
||||
}
|
||||
self.apply_power_limits("retry after a refusal");
|
||||
}
|
||||
if self.running && self.node.is_some() && !self.miners.is_empty() && now.duration_since(self.merge_view_at) >= Duration::from_secs(30) {
|
||||
self.merge_view_at = now;
|
||||
let shared = self.shared.clone();
|
||||
|
|
@ -5334,6 +5388,17 @@ fn point_words(p: &crate::ember::Point) -> String {
|
|||
if p.clock_mhz > 0 { format!("{} MHz at {}%", p.clock_mhz, p.power_pct) } else { format!("{}% (clock unlocked)", p.power_pct) }
|
||||
}
|
||||
|
||||
/// After a cap the card did not take: when to ask again (2 min, then 10 min), and None on the third refusal, when the
|
||||
/// engine writes the FAULT line instead and waits for a setting change or the card tile (main's rule, 7 October 2026:
|
||||
/// a refused limit is re-applied or reported, never left as "cap NOT applied").
|
||||
pub fn cap_refusal_plan(refusals: u32) -> Option<Duration> {
|
||||
match refusals {
|
||||
0 | 1 => Some(Duration::from_secs(120)),
|
||||
2 => Some(Duration::from_secs(600)),
|
||||
_ => None,
|
||||
}
|
||||
}
|
||||
|
||||
fn requested_watts(c: &CardState) -> f64 {
|
||||
let pct = if c.power_pct == 0 { 80 } else { c.power_pct.clamp(crate::sweep::MIN_PCT, 100) };
|
||||
let mut w = c.power_default_w * pct as f64 / 100.0;
|
||||
|
|
@ -5528,6 +5593,15 @@ mod resume_tests {
|
|||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
|
||||
/// A refused cap is asked again at 2 and 10 minutes, then reported as a FAULT line (never left as "cap NOT applied")
|
||||
#[test]
|
||||
fn a_refused_cap_climbs_the_retry_ladder_then_faults() {
|
||||
assert_eq!(super::cap_refusal_plan(1), Some(std::time::Duration::from_secs(120)));
|
||||
assert_eq!(super::cap_refusal_plan(2), Some(std::time::Duration::from_secs(600)));
|
||||
assert_eq!(super::cap_refusal_plan(3), None, "the third refusal is the FAULT line");
|
||||
assert_eq!(super::cap_refusal_plan(7), None);
|
||||
}
|
||||
#[test]
|
||||
fn power_control_off_builds_no_elevated_command() {
|
||||
// the decision (the project lead, 5 October 2026): off = the app never asks; the elevated PC sweep job is the exception
|
||||
|
|
|
|||
|
|
@ -248,11 +248,15 @@ impl Jobs {
|
|||
active: None,
|
||||
needs_logged: std::collections::HashSet::new(),
|
||||
fingerprint: manifest::fingerprint(manifest::OTA_PUBLIC_KEY_HEX),
|
||||
wake: Arc::new(WakeCtl { on: AtomicBool::new(allowed) }),
|
||||
wake: Arc::new(WakeCtl::new(allowed, &shared.runtime.id8())),
|
||||
wake_pending: false,
|
||||
last_published: String::new(),
|
||||
};
|
||||
j.publish(shared);
|
||||
// the ping names the last job this machine ran, across restarts (the newest record in jobs-state.json)
|
||||
if let Some(last) = j.ledger.records.values().max_by_key(|r| r.started_at) {
|
||||
j.wake.set_last_job(&last.id);
|
||||
}
|
||||
if !j.url.is_empty() {
|
||||
let wake_url = wake_url_of(std::env::var("IGNEUM_APP_JOBS_WAKE_URL").ok());
|
||||
if !wake_url.is_empty() {
|
||||
|
|
@ -568,6 +572,7 @@ impl Jobs {
|
|||
return None;
|
||||
}
|
||||
shared.event("info", &format!("job {} ({}) starts: {}", job.id, job.kind, job.label()));
|
||||
self.wake.set_last_job(&job.id);
|
||||
let ctl = Arc::new(Ctl::default());
|
||||
let needs_miners_stopped = job.kind == "shard-benchmark" || (job.kind == "run" && job.bool_param("stop_miners_first"));
|
||||
let cards_off: Vec<String> = if job.kind == "run" { job.list_param("cards_off") } else { vec![] };
|
||||
|
|
@ -827,9 +832,43 @@ fn upload_file(shared: &Arc<Shared>, job: &Job, path: &Path, label_prefix: &str)
|
|||
// busy-loops: a reply that came back early is followed by the rest of a 10 s floor, a failing endpoint backs off
|
||||
// 5, 15, then 60 s, and an empty stamp (nothing published yet) waits a full hold.
|
||||
|
||||
/// Shared with the waker thread: it polls only while remote jobs are allowed.
|
||||
/// Shared with the waker thread: it polls only while remote jobs are allowed. The ping (MF-11, 0.3.21): every wake
|
||||
/// request names this machine (its id8, no secret), the app version and the last job it ran, so the relay can show a
|
||||
/// machine whose job channel has gone quiet ("silent since <time>, last job <name>") the moment 15 minutes pass with no
|
||||
/// poll; a dead app polls nothing, and before this nothing on the console said so until the next upload gap was noticed.
|
||||
pub struct WakeCtl {
|
||||
on: AtomicBool,
|
||||
machine: String,
|
||||
last_job: Mutex<String>,
|
||||
}
|
||||
|
||||
impl WakeCtl {
|
||||
pub fn new(on: bool, machine: &str) -> WakeCtl {
|
||||
WakeCtl { on: AtomicBool::new(on), machine: machine.to_string(), last_job: Mutex::new(String::new()) }
|
||||
}
|
||||
pub fn set_last_job(&self, id: &str) {
|
||||
*self.last_job.lock().unwrap() = id.to_string();
|
||||
}
|
||||
/// The query fragment of the ping: machine=<id8>&v=<version>[&job=<id>]; values are the app's own, URL-safe by shape.
|
||||
pub fn ping(&self) -> String {
|
||||
ping_query(&self.machine, crate::engine::VERSION, &self.last_job.lock().unwrap())
|
||||
}
|
||||
}
|
||||
|
||||
/// machine and job ids as the relay reads them: id8 is 8 hex; a job id is the publisher's `[\w.-]` name; anything else
|
||||
/// is dropped from the query rather than escaped (the relay clips and validates on its side too).
|
||||
pub fn ping_query(machine: &str, version: &str, last_job: &str) -> String {
|
||||
let safe = |s: &str, max: usize| -> String { s.chars().filter(|c| c.is_ascii_alphanumeric() || matches!(c, '.' | '-' | '_' | ':')).take(max).collect() };
|
||||
let m = safe(machine, 16);
|
||||
if m.is_empty() {
|
||||
return String::new();
|
||||
}
|
||||
let mut q = format!("machine={m}&v={}", safe(version, 32));
|
||||
let j = safe(last_job, 80);
|
||||
if !j.is_empty() {
|
||||
q.push_str(&format!("&job={j}"));
|
||||
}
|
||||
q
|
||||
}
|
||||
|
||||
/// The stamp logic, free of I/O for the tests.
|
||||
|
|
@ -892,17 +931,24 @@ fn wake_url_of(env: Option<String>) -> String {
|
|||
}
|
||||
}
|
||||
|
||||
fn wake_query(url: &str, since: &str) -> String {
|
||||
if since.is_empty() {
|
||||
url.to_string()
|
||||
} else {
|
||||
format!("{url}{}since={since}", if url.contains('?') { '&' } else { '?' })
|
||||
fn wake_query(url: &str, since: &str, ping: &str) -> String {
|
||||
let mut out = url.to_string();
|
||||
let mut sep = if url.contains('?') { '&' } else { '?' };
|
||||
if !since.is_empty() {
|
||||
out.push(sep);
|
||||
out.push_str(&format!("since={since}"));
|
||||
sep = '&';
|
||||
}
|
||||
if !ping.is_empty() {
|
||||
out.push(sep);
|
||||
out.push_str(ping);
|
||||
}
|
||||
out
|
||||
}
|
||||
|
||||
/// One long-poll. Ok carries the relay's stamp (empty when it holds none).
|
||||
fn wake_request(url: &str, since: &str) -> Result<String, String> {
|
||||
let full = wake_query(url, since);
|
||||
fn wake_request(url: &str, since: &str, ping: &str) -> Result<String, String> {
|
||||
let full = wake_query(url, since, ping);
|
||||
let max_time = (WAKE_HOLD_S + 13).to_string();
|
||||
let (code, out) = run_capture(Command::new(crate::platform::tool("curl")).args(["-fsS", "--max-time", &max_time, &full]), Duration::from_secs(WAKE_HOLD_S + 20));
|
||||
if code != Some(0) {
|
||||
|
|
@ -923,7 +969,7 @@ fn wake_loop(shared: Arc<Shared>, url: String, ctl: Arc<WakeCtl>) {
|
|||
continue;
|
||||
}
|
||||
let t0 = Instant::now();
|
||||
match wake_request(&url, &st.stamp) {
|
||||
match wake_request(&url, &st.stamp, &ctl.ping()) {
|
||||
Ok(stamp) => {
|
||||
let (r, recovered) = st.reply(&stamp);
|
||||
if recovered {
|
||||
|
|
@ -1690,9 +1736,16 @@ mod tests {
|
|||
assert_eq!(wake_url_of(None), WAKE_URL);
|
||||
assert_eq!(wake_url_of(Some(String::new())), "");
|
||||
assert_eq!(wake_url_of(Some(" http://127.0.0.1:4180/wake ".into())), "http://127.0.0.1:4180/wake");
|
||||
assert_eq!(wake_query("https://r/wake", ""), "https://r/wake");
|
||||
assert_eq!(wake_query("https://r/wake", "2026-10-05T11:02:17Z.5e7b56f5"), "https://r/wake?since=2026-10-05T11:02:17Z.5e7b56f5");
|
||||
assert_eq!(wake_query("https://r/api/wake?x=1", "S"), "https://r/api/wake?x=1&since=S");
|
||||
assert_eq!(wake_query("https://r/wake", "", ""), "https://r/wake");
|
||||
assert_eq!(wake_query("https://r/wake", "2026-10-05T11:02:17Z.5e7b56f5", ""), "https://r/wake?since=2026-10-05T11:02:17Z.5e7b56f5");
|
||||
assert_eq!(wake_query("https://r/api/wake?x=1", "S", ""), "https://r/api/wake?x=1&since=S");
|
||||
// the ping (MF-11): the machine, the version and the last job ride on every wake request
|
||||
assert_eq!(wake_query("https://r/wake", "", "machine=1ccfe586&v=0.3.21"), "https://r/wake?machine=1ccfe586&v=0.3.21");
|
||||
assert_eq!(wake_query("https://r/wake", "S", "machine=1ccfe586&v=0.3.21&job=update-now-0319"), "https://r/wake?since=S&machine=1ccfe586&v=0.3.21&job=update-now-0319");
|
||||
assert_eq!(ping_query("1ccfe586", "0.3.21", ""), "machine=1ccfe586&v=0.3.21");
|
||||
assert_eq!(ping_query("1ccfe586", "0.3.21", "update-now-0319-1ccfe586"), "machine=1ccfe586&v=0.3.21&job=update-now-0319-1ccfe586");
|
||||
assert_eq!(ping_query("", "0.3.21", "x"), "", "no machine, no ping");
|
||||
assert_eq!(ping_query("1ccfe586", "0.3.21", "a b&c=d"), "machine=1ccfe586&v=0.3.21&job=abcd", "only the safe characters travel");
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -49,6 +49,7 @@ mod chainfacts;
|
|||
mod card;
|
||||
mod drivertable;
|
||||
mod drivers;
|
||||
mod bootcheck;
|
||||
|
||||
use std::io::{BufRead, Write};
|
||||
use std::sync::mpsc::channel;
|
||||
|
|
|
|||
|
|
@ -380,6 +380,12 @@ impl Updater {
|
|||
let rolled_back = v.get("rolled_back").and_then(|x| x.as_bool()).unwrap_or(false);
|
||||
let deferred = v.get("deferred").and_then(|x| x.as_bool()).unwrap_or(false);
|
||||
let ver = v.get("version").and_then(|x| x.as_str()).unwrap_or("").to_string();
|
||||
// 0.3.21: the helper says how the app came back (ok, rolled-back, relaunched, installer-failed) and after how long
|
||||
let ret = v.get("return").and_then(|x| x.as_str()).unwrap_or("");
|
||||
let ready_s = v.get("ready_s").and_then(|x| x.as_i64()).unwrap_or(-1);
|
||||
if !ret.is_empty() {
|
||||
shared.log(&format!("update-return: the helper reports '{ret}' for {ver}{}", if ready_s >= 0 { format!(", an engine answered after {ready_s} s") } else { ", no engine answered inside its window".to_string() }));
|
||||
}
|
||||
if !ok && deferred {
|
||||
// the installer never ran (nobody answered the administrator prompt): not a failure, it tries again
|
||||
shared.log(&format!("OTA: the update to {ver} was deferred before this start ({err}); it tries again in this machine's slot"));
|
||||
|
|
@ -426,6 +432,12 @@ impl Updater {
|
|||
self.pending.as_ref().map(|p| p.starts >= 3).unwrap_or(false)
|
||||
}
|
||||
|
||||
/// The version this engine replaced, on the first start after an update (MF-11: the read-back line the engine logs
|
||||
/// as its first act, "app <version> up after the update from <from>"); None on any other start.
|
||||
pub fn updated_from(&self) -> Option<String> {
|
||||
self.pending.as_ref().filter(|p| p.starts == 1 && p.to == self.current).map(|p| p.from.clone())
|
||||
}
|
||||
|
||||
// ---- state for the dashboard -------------------------------------------------------------------------------------
|
||||
|
||||
fn publish(&self, shared: &Arc<Shared>) {
|
||||
|
|
@ -951,6 +963,10 @@ impl Updater {
|
|||
c.args(["-NoProfile", "-ExecutionPolicy", "Bypass", "-WindowStyle", "Hidden", "-File"]).arg(&script).args([
|
||||
"-Mode", "apply", "-EnginePid", &std::process::id().to_string(), "-Installer", &staged.display().to_string(), "-Version", &to, "-Result", &result.display().to_string(), "-InstallDir", &install_dir.display().to_string(), "-Sha256", &entry.sha256,
|
||||
]);
|
||||
// 0.3.21 (MF-11): the helper owns the return. It keeps the exe set beside the app, launches the app itself after
|
||||
// the installer, polls the new engine's api/state, restores the kept set when nothing answers, and posts one line
|
||||
// to the intake either way (the key it needs is in igneum-app.json beside the exe; nothing on the command line).
|
||||
c.args(["-Previous", &previous_dir_for(&install_dir).display().to_string(), "-Machine", &shared.runtime.id8(), "-Intake", &shared.packaged.log_intake_url, "-AppDir", &self.app_dir.display().to_string()]);
|
||||
let per_user = !under_program_files(&install_dir);
|
||||
shared.log(&format!("update: starting the installer first, the miners keep running: {} ({})", script.display(), if per_user { "per-user install, no administrator prompt" } else { "install in Program Files: Windows asks for administrator approval" }));
|
||||
if !per_user {
|
||||
|
|
@ -1002,6 +1018,7 @@ impl Updater {
|
|||
let mut c = Command::new(crate::platform::tool("powershell"));
|
||||
c.args(["-NoProfile", "-ExecutionPolicy", "Bypass", "-WindowStyle", "Hidden", "-File"]).arg(&script).args([
|
||||
"-Sha256", &sha, "-Mode", "rollback", "-EnginePid", &std::process::id().to_string(), "-Installer", &p.previous_installer, "-Version", &p.to, "-Result", &result.display().to_string(), "-InstallDir", &install_dir.display().to_string(),
|
||||
"-Previous", &previous_dir_for(&install_dir).display().to_string(), "-Machine", &shared.runtime.id8(), "-Intake", &shared.packaged.log_intake_url, "-AppDir", &self.app_dir.display().to_string(),
|
||||
]);
|
||||
spawn_detached(&mut c)?;
|
||||
Ok(())
|
||||
|
|
@ -1089,6 +1106,189 @@ fn installer_name_for(version: &str) -> String {
|
|||
format!("Igneum-Miner-Setup-{version}.exe")
|
||||
}
|
||||
|
||||
/// Windows: the folder the helper keeps the running exe set in before the installer runs, beside the app
|
||||
/// (`<install dir>.previous`, so `...\Programs\Igneum Miner.previous`). A rollback copies it back over the install folder.
|
||||
#[allow(dead_code)]
|
||||
fn previous_dir_for(install_dir: &Path) -> PathBuf {
|
||||
let name = install_dir.file_name().map(|n| n.to_string_lossy().into_owned()).unwrap_or_else(|| "Igneum Miner".into());
|
||||
install_dir.with_file_name(format!("{name}.previous"))
|
||||
}
|
||||
|
||||
// ---- the return after a Windows install (MF-11, 0.3.21) ------------------------------------------------------------
|
||||
//
|
||||
// PC 2 took the 0.3.19 update-now at 10:34Z on 7 October 2026 and was silent from 10:46Z. Until 0.3.21 the Windows helper's
|
||||
// job ended when the installer exited 0: the installer's own [Run] entry relaunched the app, nobody checked that an engine
|
||||
// answered, the window host never restarted an engine that died, and a second launch deferred to a surviving host through
|
||||
// its single-instance mutex. The sequence below is what the helper does from the installer's exit on, written once here
|
||||
// so a test can drive it with injected exit codes and answers, and mirrored line for line by WIN_HELPER's PowerShell.
|
||||
|
||||
/// How long the helper waits for an engine to answer api/state after a launch, and how often it asks.
|
||||
pub const RETURN_READY_S: u64 = 120;
|
||||
pub const RETURN_POLL_S: u64 = 3;
|
||||
|
||||
/// One step of the helper after the installer exits.
|
||||
#[derive(Debug, Clone, PartialEq)]
|
||||
pub enum ReturnStep {
|
||||
/// start igneum-app.exe --launch from the install folder (a detached process; the helper outlives the old engine)
|
||||
Launch,
|
||||
/// poll app.url + api/state for RETURN_READY_S; `want` is the version that must answer ("" = any engine)
|
||||
WaitReady { want: String },
|
||||
/// quit through the API, then end what is left by name (the installer's own stop order)
|
||||
StopAll,
|
||||
/// copy the kept exe set (`<install dir>.previous`) back over the install folder
|
||||
RestorePrevious,
|
||||
/// the result file for the next engine and the one intake line
|
||||
Done { ok: bool, rolled_back: bool, fault: bool, how: &'static str },
|
||||
}
|
||||
|
||||
/// What a WaitReady step saw: the version that answered, or nobody.
|
||||
pub type Answer = Option<String>;
|
||||
|
||||
/// The helper's sequence from the installer's exit code on. `answers` is consulted once per WaitReady, in order (the test
|
||||
/// injects them; the PowerShell asks the engine). `previous_kept` says whether the exe set was copied aside before the
|
||||
/// installer ran.
|
||||
pub fn return_sequence(installer_exit: i32, version: &str, previous_kept: bool, mut answers: impl FnMut(usize) -> Answer) -> Vec<ReturnStep> {
|
||||
let mut steps = vec![ReturnStep::Launch];
|
||||
let want = if installer_exit == 0 { version.to_string() } else { String::new() };
|
||||
steps.push(ReturnStep::WaitReady { want: want.clone() });
|
||||
match answers(0) {
|
||||
Some(v) if want.is_empty() || v == want => {
|
||||
steps.push(if installer_exit == 0 { ReturnStep::Done { ok: true, rolled_back: false, fault: false, how: "ok" } } else { ReturnStep::Done { ok: false, rolled_back: false, fault: true, how: "installer-failed" } });
|
||||
return steps;
|
||||
}
|
||||
_ => {}
|
||||
}
|
||||
// nothing (or the wrong engine) answered inside the window: back to the kept version
|
||||
steps.push(ReturnStep::StopAll);
|
||||
if previous_kept {
|
||||
steps.push(ReturnStep::RestorePrevious);
|
||||
steps.push(ReturnStep::Launch);
|
||||
steps.push(ReturnStep::WaitReady { want: String::new() });
|
||||
let how = if answers(1).is_some() { "rolled-back" } else { "rolled-back-silent" };
|
||||
steps.push(ReturnStep::Done { ok: false, rolled_back: true, fault: true, how });
|
||||
} else {
|
||||
steps.push(ReturnStep::Launch);
|
||||
steps.push(ReturnStep::Done { ok: false, rolled_back: false, fault: true, how: "relaunched" });
|
||||
}
|
||||
steps
|
||||
}
|
||||
|
||||
/// The helper before 0.3.21, for the record: the installer's exit code alone decided the result and nothing was asked
|
||||
/// of the new engine (the known-failed shape of MF-11).
|
||||
pub fn legacy_return_sequence(installer_exit: i32) -> Vec<ReturnStep> {
|
||||
if installer_exit == 0 { vec![ReturnStep::Done { ok: true, rolled_back: false, fault: false, how: "" }] } else { vec![ReturnStep::Launch, ReturnStep::Done { ok: false, rolled_back: false, fault: false, how: "" }] }
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod return_tests {
|
||||
use super::*;
|
||||
fn done(steps: &[ReturnStep]) -> &ReturnStep {
|
||||
steps.last().unwrap()
|
||||
}
|
||||
fn answers(list: &[Answer]) -> impl FnMut(usize) -> Answer + '_ {
|
||||
move |i| list.get(i).cloned().flatten()
|
||||
}
|
||||
|
||||
/// Known-failed first: the helper before 0.3.21 reported ok on the installer's exit 0 with nobody answering.
|
||||
#[test]
|
||||
fn the_legacy_helper_reports_ok_with_no_engine_up() {
|
||||
let steps = legacy_return_sequence(0);
|
||||
assert_eq!(steps, vec![ReturnStep::Done { ok: true, rolled_back: false, fault: false, how: "" }]);
|
||||
assert!(!steps.iter().any(|s| matches!(s, ReturnStep::WaitReady { .. })), "nothing was asked of the new engine");
|
||||
assert!(!steps.iter().any(|s| matches!(s, ReturnStep::Launch)), "the launch was the installer's, not the helper's");
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn installer_ok_and_the_new_engine_answers_is_ok() {
|
||||
let steps = return_sequence(0, "0.3.21", true, answers(&[Some("0.3.21".into())]));
|
||||
assert_eq!(steps, vec![ReturnStep::Launch, ReturnStep::WaitReady { want: "0.3.21".into() }, ReturnStep::Done { ok: true, rolled_back: false, fault: false, how: "ok" }]);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn installer_ok_and_nobody_answers_restores_the_previous_exe_set() {
|
||||
let steps = return_sequence(0, "0.3.21", true, answers(&[None, Some("0.3.20".into())]));
|
||||
assert_eq!(steps, vec![
|
||||
ReturnStep::Launch, ReturnStep::WaitReady { want: "0.3.21".into() }, ReturnStep::StopAll, ReturnStep::RestorePrevious, ReturnStep::Launch, ReturnStep::WaitReady { want: String::new() },
|
||||
ReturnStep::Done { ok: false, rolled_back: true, fault: true, how: "rolled-back" },
|
||||
]);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn the_old_engine_still_answering_after_exit_0_is_not_the_new_one() {
|
||||
// the installer said 0 but never replaced the running engine (files in use): the old version answers, the window
|
||||
// ends in a rollback to the kept set, which is the same version, and a FAULT line says so
|
||||
let steps = return_sequence(0, "0.3.21", true, answers(&[Some("0.3.20".into()), Some("0.3.20".into())]));
|
||||
assert!(matches!(done(&steps), ReturnStep::Done { ok: false, rolled_back: true, fault: true, .. }));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn nobody_answers_twice_is_still_reported() {
|
||||
let steps = return_sequence(0, "0.3.21", true, answers(&[None, None]));
|
||||
assert_eq!(*done(&steps), ReturnStep::Done { ok: false, rolled_back: true, fault: true, how: "rolled-back-silent" });
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn without_a_kept_set_the_helper_relaunches_what_is_there_and_reports() {
|
||||
let steps = return_sequence(0, "0.3.21", false, answers(&[None]));
|
||||
assert_eq!(steps, vec![ReturnStep::Launch, ReturnStep::WaitReady { want: "0.3.21".into() }, ReturnStep::StopAll, ReturnStep::Launch, ReturnStep::Done { ok: false, rolled_back: false, fault: true, how: "relaunched" }]);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn a_failed_installer_relaunches_the_old_version_and_reports_a_fault() {
|
||||
// exit 5 (cancelled) or 8 (files in use, a restart wanted): any engine answering is the old one, kept, with a FAULT line
|
||||
for code in [1, 5, 8] {
|
||||
let steps = return_sequence(code, "0.3.21", true, answers(&[Some("0.3.20".into())]));
|
||||
assert_eq!(steps, vec![ReturnStep::Launch, ReturnStep::WaitReady { want: String::new() }, ReturnStep::Done { ok: false, rolled_back: false, fault: true, how: "installer-failed" }], "exit {code}");
|
||||
}
|
||||
let steps = return_sequence(5, "0.3.21", true, answers(&[None, Some("0.3.20".into())]));
|
||||
assert!(matches!(done(&steps), ReturnStep::Done { ok: false, rolled_back: true, fault: true, .. }));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn every_sequence_launches_before_it_waits_and_ends_in_a_done() {
|
||||
for code in [0, 1, 5, 8] {
|
||||
for kept in [true, false] {
|
||||
for a in [vec![None, None], vec![Some("0.3.21".to_string()), None], vec![None, Some("0.3.20".to_string())]] {
|
||||
let steps = return_sequence(code, "0.3.21", kept, answers(&a));
|
||||
assert_eq!(steps[0], ReturnStep::Launch);
|
||||
assert!(matches!(steps[1], ReturnStep::WaitReady { .. }));
|
||||
assert!(matches!(done(&steps), ReturnStep::Done { .. }));
|
||||
let fault = matches!(done(&steps), ReturnStep::Done { fault: true, .. });
|
||||
let ok = matches!(done(&steps), ReturnStep::Done { ok: true, .. });
|
||||
assert!(ok != fault, "a result is ok or a fault, never neither: {steps:?}");
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// The PowerShell mirrors the sequence: every step has its line, in order, and the installer is told the helper relaunches.
|
||||
#[test]
|
||||
fn the_windows_helper_carries_every_step() {
|
||||
let h = WIN_HELPER;
|
||||
let at = |s: &str| h.find(s).unwrap_or_else(|| panic!("WIN_HELPER lacks '{s}'"));
|
||||
assert!(at("/IGNOTA=2") > 0, "the installer must not relaunch (IGNOTA=1 is the old helpers' path)");
|
||||
assert!(h.contains(&format!("$ReadyS = {RETURN_READY_S}")) && h.contains(&format!("$PollS = {RETURN_POLL_S}")));
|
||||
let robocopy_keep = at("robocopy $InstallDir $Previous");
|
||||
let installer = at("Start-Process -FilePath $Installer"); // console: a test marker, not a spawn (the helper line above it carries the comment)
|
||||
let launch = at("function Launch()");
|
||||
let wait = at("function WaitReady(");
|
||||
let stop = at("function StopAll()");
|
||||
let restore = at("robocopy $Previous $InstallDir");
|
||||
let report = at("function Report(");
|
||||
assert!(launch < installer && wait < installer && stop < installer && report < installer, "the functions are defined before the installer runs");
|
||||
assert!(robocopy_keep < installer && restore < installer, "the keep and the restore are functions defined before the installer line");
|
||||
assert!(at("$kept = KeepPrevious") < installer, "the exe set is kept before the installer runs");
|
||||
assert!(at("Comeback $code $kept") > installer, "the return runs after the installer");
|
||||
for marker in ["'ok'", "'installer-failed'", "'rolled-back'", "'rolled-back-silent'", "'relaunched'"] {
|
||||
assert!(h.contains(marker), "the helper never writes return={marker}");
|
||||
}
|
||||
assert!(h.contains("FAULT update-return:"), "the fault line");
|
||||
assert!(h.contains("update-return: ok"), "the ok line");
|
||||
assert!(h.contains("api/state"), "readiness is the engine's own answer");
|
||||
assert!(!h.contains("-IntakeKey"), "no key travels on a command line (R4.3.8)");
|
||||
}
|
||||
}
|
||||
|
||||
fn curl(args: &[&str], limit: Duration) -> Result<(), String> {
|
||||
let mut c = Command::new(crate::platform::tool("curl"));
|
||||
c.args(args);
|
||||
|
|
@ -1324,60 +1524,180 @@ case "$MODE" in
|
|||
esac
|
||||
"#;
|
||||
|
||||
#[cfg(windows)]
|
||||
/// The Windows helper. Not cfg-gated so the return test above can read it on every platform.
|
||||
#[allow(dead_code)]
|
||||
const WIN_HELPER: &str = r#"# Igneum Miner update helper, written by the engine (src/ota.rs). Not for running by hand.
|
||||
# powershell -File ota-apply.ps1 -Mode apply|rollback -EnginePid <pid> -Installer <setup exe> -Version <v> -Result <json> -InstallDir <folder> -Sha256 <hex>
|
||||
# powershell -File ota-apply.ps1 -Mode apply|rollback -EnginePid <pid> -Installer <setup exe> -Version <v> -Result <json> -InstallDir <folder> -Sha256 <hex> -Previous <folder> -Machine <id8> -Intake <url> -AppDir <app data>
|
||||
# The installer runs FIRST, while the engine keeps mining (4 October 2026: two unattended PCs sat stopped at an
|
||||
# administrator prompt nobody could click). A per-user installer (0.3.3 and later, PrivilegesRequired=lowest) needs no
|
||||
# prompt; an older administrator installer raises one through ShellExecute. Only when the installer actually runs does
|
||||
# its PrepareToInstall step stop the engine (api/quit: miners first, then the node), replace the files and relaunch
|
||||
# the app (/IGNOTA=1). A declined, timed-out or unanswered prompt leaves the engine running: the result says
|
||||
# deferred:true and the engine shows "waits for the next time someone is at this PC". The old app is relaunched only
|
||||
# when the engine is gone and the install did not happen.
|
||||
param([string]$Mode, [int]$EnginePid, [string]$Installer, [string]$Version, [string]$Result, [string]$InstallDir, [string]$Sha256 = '')
|
||||
# its PrepareToInstall step stop the engine (api/quit: miners first, then the node) and replace the files. A declined,
|
||||
# timed-out or unanswered prompt leaves the engine running: the result says deferred:true and the engine shows "waits for
|
||||
# the next time someone is at this PC".
|
||||
# From 0.3.21 (MF-11, 7 October 2026) this helper owns the return: it keeps the running exe set beside the app before the
|
||||
# installer runs, starts the app itself afterwards (/IGNOTA=2 tells the installer not to), waits for an engine to answer
|
||||
# api/state with the new version, restores the kept set when nothing answers inside the window, and posts one line to
|
||||
# the log intake either way (the key is read from igneum-app.json beside the exe, never from the command line). The
|
||||
# sequence is src/ota.rs return_sequence(), tested there with injected exit codes; this file mirrors it step for step.
|
||||
param([string]$Mode, [int]$EnginePid, [string]$Installer, [string]$Version, [string]$Result, [string]$InstallDir, [string]$Sha256 = '', [string]$Previous = '', [string]$Machine = '', [string]$Intake = '', [string]$AppDir = '')
|
||||
$log = Join-Path (Split-Path -Parent $Result) 'ota-apply.log'
|
||||
if (-not $AppDir) { $AppDir = Split-Path -Parent $Result }
|
||||
$ReadyS = 120
|
||||
$PollS = 3
|
||||
function Log([string]$t) { Add-Content -Path $log -Value ("{0} {1}" -f (Get-Date -Format s), $t) }
|
||||
function Done([bool]$ok, [string]$err, [bool]$rb, [bool]$deferred) {
|
||||
$o = @{ ok = $ok; version = $Version; error = $err; rolled_back = $rb; deferred = $deferred; at = [int][double](Get-Date -UFormat %s) }
|
||||
function Done([bool]$ok, [string]$err, [bool]$rb, [bool]$deferred, [string]$ret, [int]$readyS) {
|
||||
$o = @{ ok = $ok; version = $Version; error = $err; rolled_back = $rb; deferred = $deferred; at = [int][double](Get-Date -UFormat %s); 'return' = $ret; ready_s = $readyS }
|
||||
($o | ConvertTo-Json -Compress) | Set-Content -Path $Result -Encoding ASCII
|
||||
}
|
||||
function EngineAlive() { return [bool](Get-Process -Id $EnginePid -ErrorAction SilentlyContinue) }
|
||||
function Relaunch() {
|
||||
if (EngineAlive) { return }
|
||||
$exe = Join-Path $InstallDir 'igneum-app.exe'
|
||||
# console: igneum-app.exe is a windows-subsystem program (no console); -WindowStyle Hidden would hide the window host it opens
|
||||
if (Test-Path $exe) { Log 'engine gone and nothing installed: starting the old app again'; Start-Process -FilePath $exe -ArgumentList '--launch' -WorkingDirectory $InstallDir | Out-Null }
|
||||
function AppUrl() {
|
||||
$f = Join-Path $AppDir 'app.url'
|
||||
if (Test-Path $f) { return (Get-Content $f -Raw).Trim() }
|
||||
return ''
|
||||
}
|
||||
Log "$Mode : engine $EnginePid installer '$Installer' version $Version (the engine keeps mining until the installer runs)"
|
||||
if (-not (Test-Path $Installer)) { Log 'installer missing'; Done $false 'the downloaded installer is missing' $false $false; exit 1 }
|
||||
function AppVersion() {
|
||||
# the engine's own answer: GET <app.url>api/state and its version field (the URL file is rewritten by every start)
|
||||
$u = AppUrl
|
||||
if (-not $u) { return '' }
|
||||
try { $r = Invoke-RestMethod -Uri ($u + 'api/state') -TimeoutSec 5 -UseBasicParsing; return [string]$r.version } catch { return '' }
|
||||
}
|
||||
function WaitReady([string]$want, [int]$limitS) {
|
||||
# the seconds until an engine answered with the wanted version (any version when $want is empty); -1 when none did
|
||||
$t0 = Get-Date
|
||||
while (((Get-Date) - $t0).TotalSeconds -lt $limitS) {
|
||||
$v = AppVersion
|
||||
if ($v -and (($want -eq '') -or ($v -eq $want))) { return [int]((Get-Date) - $t0).TotalSeconds }
|
||||
Start-Sleep -Seconds $PollS
|
||||
}
|
||||
return -1
|
||||
}
|
||||
function Launch() {
|
||||
$exe = Join-Path $InstallDir 'igneum-app.exe'
|
||||
if (-not (Test-Path $exe)) { Log ("nothing to start: " + $exe + " is missing"); return $false }
|
||||
Log ("starting " + $exe + " --launch")
|
||||
# console: igneum-app.exe is a windows-subsystem program (no console); -WindowStyle Hidden would hide the window host it opens
|
||||
Start-Process -FilePath $exe -ArgumentList '--launch' -WorkingDirectory $InstallDir | Out-Null
|
||||
return $true
|
||||
}
|
||||
function StopAll() {
|
||||
# the quit through the API first (miners, then the node), then whatever is left by name: the installer's own order
|
||||
$u = AppUrl
|
||||
if ($u) { try { Invoke-WebRequest -Uri ($u + 'api/quit') -Method POST -Body '{}' -ContentType 'application/json' -UseBasicParsing -TimeoutSec 5 | Out-Null } catch { } }
|
||||
$deadline = (Get-Date).AddSeconds(40)
|
||||
while ((Get-Date) -lt $deadline) {
|
||||
if (@(Get-Process -Name 'igneum-app', 'igneumd' -ErrorAction SilentlyContinue).Count -eq 0) { break }
|
||||
Start-Sleep -Milliseconds 500
|
||||
}
|
||||
foreach ($n in @('Igneum Miner', 'igneum-app', 'igneum-miner', 'igneum-worker-cuda', 'igneum-worker-opencl', 'igneumd')) {
|
||||
Get-Process -Name $n -ErrorAction SilentlyContinue | ForEach-Object { Stop-Process -Id $_.Id -Force -ErrorAction SilentlyContinue }
|
||||
}
|
||||
Start-Sleep -Seconds 1
|
||||
}
|
||||
function Report([string]$line) {
|
||||
# one line to the log intake, label fault-win-<id8>, as site/api/log.mjs expects; the key is the one the installed
|
||||
# app carries (igneum-app.json beside the exe, or the kept copy); nothing to post with is logged, never fatal
|
||||
Log $line
|
||||
if (-not $Intake -or -not $Machine) { return }
|
||||
$cfg = Join-Path $InstallDir 'igneum-app.json'
|
||||
if (-not (Test-Path $cfg) -and $Previous) { $cfg = Join-Path $Previous 'igneum-app.json' }
|
||||
if (-not (Test-Path $cfg)) { Log 'no igneum-app.json to read the intake key from; the line stays in this log'; return }
|
||||
try {
|
||||
$key = [string](Get-Content $cfg -Raw | ConvertFrom-Json).log_intake_key
|
||||
if (-not $key) { Log 'igneum-app.json carries no intake key'; return }
|
||||
$o = @{ label = ('fault-win-' + $Machine); machine = ($env:COMPUTERNAME + '-' + $Machine); run_id = ('update-return-' + $Version); lines = ("IGNEUM-APP version=" + $Version + " machine=" + $Machine + " platform=windows node=?`n" + $line) }
|
||||
$bytes = [Text.Encoding]::UTF8.GetBytes(($o | ConvertTo-Json -Compress))
|
||||
Invoke-RestMethod -Method Post -Uri $Intake -Headers @{ 'x-igneum-key' = $key } -ContentType 'application/json; charset=utf-8' -Body $bytes -TimeoutSec 30 | Out-Null
|
||||
Log 'intake line posted'
|
||||
} catch { Log ('intake post failed: ' + $_.Exception.Message) }
|
||||
}
|
||||
function KeepPrevious() {
|
||||
# the running exe set beside the app, what a rollback restores; packs, build and dist are rebuilt or unneeded
|
||||
if (-not $Previous) { return $false }
|
||||
try {
|
||||
& robocopy $InstallDir $Previous /MIR /XD packs build dist /R:1 /W:1 /NFL /NDL /NJH /NJS /NP | Out-Null
|
||||
if ($LASTEXITCODE -lt 8 -and (Test-Path (Join-Path $Previous 'igneum-app.exe'))) { Log ("previous version kept at " + $Previous); return $true }
|
||||
Log ("robocopy could not keep the previous version (exit " + $LASTEXITCODE + ")")
|
||||
} catch { Log ('could not keep the previous version: ' + $_.Exception.Message) }
|
||||
return $false
|
||||
}
|
||||
function RestorePrevious() {
|
||||
if (-not $Previous -or -not (Test-Path (Join-Path $Previous 'igneum-app.exe'))) { return $false }
|
||||
try {
|
||||
& robocopy $Previous $InstallDir /MIR /XD packs build dist /R:2 /W:2 /NFL /NDL /NJH /NJS /NP | Out-Null
|
||||
if ($LASTEXITCODE -lt 8) { Log 'previous version restored over the install folder'; return $true }
|
||||
Log ("robocopy could not restore the previous version (exit " + $LASTEXITCODE + ")")
|
||||
} catch { Log ('could not restore the previous version: ' + $_.Exception.Message) }
|
||||
return $false
|
||||
}
|
||||
function Comeback([int]$code, [bool]$kept) {
|
||||
# src/ota.rs return_sequence(): Launch, WaitReady, then Done or StopAll, RestorePrevious, Launch, WaitReady, Done
|
||||
if (-not (EngineAlive)) { Launch | Out-Null } else { Log 'the engine is still up after the installer (it did not stop it)' }
|
||||
$want = ''
|
||||
if ($code -eq 0) { $want = $Version }
|
||||
$ready = WaitReady $want $ReadyS
|
||||
if ($ready -ge 0) {
|
||||
if ($code -eq 0) {
|
||||
Done $true '' $false $false 'ok' $ready
|
||||
Report ("update-return: ok " + $Version + " up in " + $ready + " s")
|
||||
exit 0
|
||||
}
|
||||
Done $false ("the installer exited with code " + $code + " (see ota-setup.log); the previous version answers again") $false $false 'installer-failed' $ready
|
||||
Report ("FAULT update-return: the installer of " + $Version + " exited with code " + $code + "; the previous version answered again after " + $ready + " s")
|
||||
exit 1
|
||||
}
|
||||
Log ("no engine answered api/state with '" + $want + "' inside " + $ReadyS + " s; back to the previous version")
|
||||
StopAll
|
||||
if ($kept -and (RestorePrevious)) {
|
||||
Launch | Out-Null
|
||||
$r2 = WaitReady '' $ReadyS
|
||||
if ($r2 -ge 0) {
|
||||
Done $false ("Igneum Miner " + $Version + " did not answer within " + $ReadyS + " s; the previous version was restored") $true $false 'rolled-back' $r2
|
||||
Report ("FAULT update-return: " + $Version + " did not answer inside " + $ReadyS + " s after the install; the previous exe set was restored and answers after " + $r2 + " s")
|
||||
exit 1
|
||||
}
|
||||
Done $false ("Igneum Miner " + $Version + " did not answer within " + $ReadyS + " s; the previous version was restored but did not answer either") $true $false 'rolled-back-silent' -1
|
||||
Report ("FAULT update-return: " + $Version + " did not answer inside " + $ReadyS + " s; the previous exe set was restored and did not answer inside " + $ReadyS + " s either; a hand start is needed on this PC")
|
||||
exit 1
|
||||
}
|
||||
Launch | Out-Null
|
||||
Done $false ("Igneum Miner " + $Version + " did not answer within " + $ReadyS + " s; no kept version to restore; what is installed was started again") $false $false 'relaunched' -1
|
||||
Report ("FAULT update-return: " + $Version + " did not answer inside " + $ReadyS + " s and no previous exe set was kept; what is installed was started again")
|
||||
exit 1
|
||||
}
|
||||
Log "$Mode : engine $EnginePid installer '$Installer' version $Version previous '$Previous' (the engine keeps mining until the installer runs)"
|
||||
if (-not (Test-Path $Installer)) { Log 'installer missing'; Done $false 'the downloaded installer is missing' $false $false '' -1; exit 1 }
|
||||
# the installer is hashed again right before it runs (R4.3.5)
|
||||
if (-not $Sha256) { Log 'no sha256 given'; Done $false 'no sha256 for the installer; not run' $false $false; exit 1 }
|
||||
if (-not $Sha256) { Log 'no sha256 given'; Done $false 'no sha256 for the installer; not run' $false $false '' -1; exit 1 }
|
||||
$have = (Get-FileHash -Path $Installer -Algorithm SHA256).Hash.ToLower()
|
||||
if ($have -ne $Sha256.ToLower()) { Log "sha256 mismatch: $have"; Remove-Item -Path $Installer -Force -ErrorAction SilentlyContinue; Done $false 'the installer changed since it was verified; not run' $false $false; exit 1 }
|
||||
if ($have -ne $Sha256.ToLower()) { Log "sha256 mismatch: $have"; Remove-Item -Path $Installer -Force -ErrorAction SilentlyContinue; Done $false 'the installer changed since it was verified; not run' $false $false '' -1; exit 1 }
|
||||
Log 'installer sha256 verified'
|
||||
$kept = $false
|
||||
if ($Mode -eq 'apply') { $kept = KeepPrevious }
|
||||
$setupLog = Join-Path (Split-Path -Parent $Result) 'ota-setup.log'
|
||||
$setupArgs = @('/VERYSILENT', '/SUPPRESSMSGBOXES', '/NORESTART', '/CLOSEAPPLICATIONS', '/IGNOTA=1', ('/LOG="' + $setupLog + '"'))
|
||||
$setupArgs = @('/VERYSILENT', '/SUPPRESSMSGBOXES', '/NORESTART', '/CLOSEAPPLICATIONS', '/IGNOTA=2', ('/LOG="' + $setupLog + '"'))
|
||||
$code = -1
|
||||
try {
|
||||
# no -Verb RunAs: a per-user installer just runs; an administrator installer makes Windows ask, and a declined or
|
||||
# timed-out prompt comes back here as an exception with the engine still mining
|
||||
# console: the Inno Setup installer is a GUI program (no console), /VERYSILENT shows nothing
|
||||
$p = Start-Process -FilePath $Installer -ArgumentList $setupArgs -Wait -PassThru
|
||||
if ($p.ExitCode -eq 0) {
|
||||
if ($Mode -eq 'rollback') { Done $false "Igneum Miner $Version did not stay up twice; the previous version was reinstalled" $true $false }
|
||||
else { Done $true '' $false $false }
|
||||
Log 'installer exit 0'
|
||||
exit 0
|
||||
}
|
||||
Log ("installer exit " + $p.ExitCode)
|
||||
Done $false ("the installer exited with code " + $p.ExitCode + " (see ota-setup.log)") $false $false
|
||||
Relaunch
|
||||
exit 1
|
||||
$code = $p.ExitCode
|
||||
} catch {
|
||||
$msg = $_.Exception.Message
|
||||
Log ("installer did not run: " + $msg)
|
||||
Log 'OTA: waiting for administrator approval; the engine keeps mining; the update waits for the next time someone is at this PC'
|
||||
Done $false ("waiting for administrator approval (" + $msg + ")") $false $true
|
||||
Relaunch
|
||||
Done $false ("waiting for administrator approval (" + $msg + ")") $false $true '' -1
|
||||
if (-not (EngineAlive)) { Launch | Out-Null }
|
||||
exit 1
|
||||
}
|
||||
Log ("installer exit " + $code)
|
||||
if ($Mode -eq 'rollback') {
|
||||
# the kept installer of the previous version ran: the same return, reported as the rollback it is
|
||||
if (-not (EngineAlive)) { Launch | Out-Null }
|
||||
$r = WaitReady '' $ReadyS
|
||||
Done $false ("Igneum Miner " + $Version + " did not stay up twice; the previous version was reinstalled") $true $false 'rolled-back' $r
|
||||
Report ("FAULT update-return: " + $Version + " did not stay up twice; the previous version was reinstalled (installer exit " + $code + ", answered after " + $r + " s)")
|
||||
exit 1
|
||||
}
|
||||
Comeback $code $kept
|
||||
"#;
|
||||
|
|
|
|||
|
|
@ -13,7 +13,7 @@ pub fn tool(name: &str) -> PathBuf {
|
|||
let sys = format!("{root}\\System32");
|
||||
let p = match name {
|
||||
"powershell" => format!("{sys}\\WindowsPowerShell\\v1.0\\powershell.exe"),
|
||||
"cmd" | "curl" | "reg" | "icacls" | "taskkill" | "w32tm" | "net" | "tar" | "wsl" => format!("{sys}\\{name}.exe"),
|
||||
"cmd" | "curl" | "reg" | "icacls" | "taskkill" | "w32tm" | "net" | "tar" | "wsl" | "wevtutil" | "robocopy" => format!("{sys}\\{name}.exe"),
|
||||
"nvidia-smi" => {
|
||||
let pf = std::env::var("ProgramFiles").unwrap_or_else(|_| "C:\\Program Files".into());
|
||||
let a = format!("{pf}\\NVIDIA Corporation\\NVSMI\\nvidia-smi.exe");
|
||||
|
|
|
|||
|
|
@ -34,6 +34,14 @@ using namespace Microsoft::WRL;
|
|||
#define ID_TRAY_PAUSE 1002
|
||||
#define ID_TRAY_QUIT 1003
|
||||
#define ID_QUIT_TIMER 7
|
||||
// MF-11 (7 October 2026): an engine that stops without a quit is started again (10 s, then 60 s after three in ten
|
||||
// minutes); a second "Igneum Miner.exe" that finds this window asks it to do so now (WM_ENGINE_RESTART), instead of
|
||||
// showing a window that says "the engine stopped" with nothing mining behind it.
|
||||
#define ID_RESTART_TIMER 8
|
||||
#define WM_ENGINE_RESTART (WM_APP + 3)
|
||||
#define RESTART_SOON_MS 10000
|
||||
#define RESTART_SLOW_MS 60000
|
||||
#define RESTART_WINDOW_MS 600000
|
||||
#define IDI_APP 1
|
||||
|
||||
static HWND g_hwnd = nullptr;
|
||||
|
|
@ -52,6 +60,8 @@ static bool g_paused = false, g_quitting = false, g_exited = false, g_webviewOk
|
|||
static NOTIFYICONDATAW g_nid = {};
|
||||
static std::wstring g_trayTitle = L"Igneum Miner";
|
||||
static ULONGLONG g_quitStarted = 0;
|
||||
static int g_restarts = 0; // engine restarts inside the current window
|
||||
static ULONGLONG g_restartWindowStart = 0; // when that window opened
|
||||
|
||||
static std::wstring widen(const std::string& s) {
|
||||
if (s.empty()) return L"";
|
||||
|
|
@ -261,6 +271,47 @@ static bool startEngine() {
|
|||
return true;
|
||||
}
|
||||
|
||||
// The engine's handles, closed before another engine is started (the reader thread has already left: it posts the
|
||||
// "gone" line only after the pipe closed).
|
||||
static void closeEngine() {
|
||||
if (g_engineIn) { CloseHandle(g_engineIn); g_engineIn = nullptr; }
|
||||
if (g_engineOut) { CloseHandle(g_engineOut); g_engineOut = nullptr; }
|
||||
if (g_engine) { CloseHandle(g_engine); g_engine = nullptr; }
|
||||
}
|
||||
|
||||
// Starts the engine again after it stopped on its own. Three restarts inside ten minutes come 10 s apart; from the
|
||||
// fourth they come a minute apart, for ever: a miner that sits stopped is a miner lost (plug, tune, play).
|
||||
static void scheduleRestart() {
|
||||
ULONGLONG now = GetTickCount64();
|
||||
if (g_restartWindowStart == 0 || now - g_restartWindowStart > RESTART_WINDOW_MS) { g_restartWindowStart = now; g_restarts = 0; }
|
||||
g_restarts++;
|
||||
UINT delay = g_restarts <= 3 ? RESTART_SOON_MS : RESTART_SLOW_MS;
|
||||
wchar_t buf[160];
|
||||
swprintf_s(buf, L"The engine stopped. Starting it again in %u s (restart %d).", delay / 1000, g_restarts);
|
||||
g_status = buf;
|
||||
setTray(L"Igneum Miner: starting the engine again");
|
||||
repaintStatus();
|
||||
SetTimer(g_hwnd, ID_RESTART_TIMER, delay, nullptr);
|
||||
}
|
||||
|
||||
static void restartEngineNow() {
|
||||
KillTimer(g_hwnd, ID_RESTART_TIMER);
|
||||
if (g_quitting || !g_exited) return;
|
||||
closeEngine();
|
||||
g_exited = false;
|
||||
g_url.clear();
|
||||
if (startEngine()) {
|
||||
g_status = L"";
|
||||
setTray(L"Igneum Miner: starting");
|
||||
repaintStatus();
|
||||
} else {
|
||||
g_exited = true;
|
||||
g_status = L"igneum-app.exe is missing next to this program. Run the installer again.";
|
||||
repaintStatus();
|
||||
scheduleRestart();
|
||||
}
|
||||
}
|
||||
|
||||
static void showTrayMenu() {
|
||||
HMENU m = CreatePopupMenu();
|
||||
AppendMenuW(m, MF_STRING, ID_TRAY_OPEN, L"Open Igneum Miner");
|
||||
|
|
@ -302,9 +353,9 @@ static LRESULT CALLBACK WndProc(HWND hwnd, UINT msg, WPARAM wp, LPARAM lp) {
|
|||
if (wp == 1) {
|
||||
g_exited = true;
|
||||
if (g_quitting) { DestroyWindow(hwnd); return 0; }
|
||||
g_status = L"The engine stopped. Close this window and open Igneum Miner again.";
|
||||
setTray(L"Igneum Miner: stopped");
|
||||
repaintStatus();
|
||||
// not a quit of ours: the engine died, or a local caller (the installer, a job) asked it to stop and nothing
|
||||
// will start it again; this window does (MF-11)
|
||||
scheduleRestart();
|
||||
return 0;
|
||||
}
|
||||
std::string* line = (std::string*)lp;
|
||||
|
|
@ -330,7 +381,21 @@ static LRESULT CALLBACK WndProc(HWND hwnd, UINT msg, WPARAM wp, LPARAM lp) {
|
|||
delete line;
|
||||
return 0;
|
||||
}
|
||||
case WM_ENGINE_RESTART:
|
||||
// a second "Igneum Miner.exe" (the update helper, the Start Menu, the relay agent's start-app) found this window:
|
||||
// an engine that is gone starts now, not in its backoff
|
||||
if (g_exited && !g_quitting) restartEngineNow();
|
||||
return 0;
|
||||
case WM_QUERYENDSESSION:
|
||||
// the Restart Manager (an installer's /CLOSEAPPLICATIONS) or a sign-out: this window closes for real, it does not
|
||||
// hide to the tray (WM_CLOSE below is the user's X)
|
||||
beginQuit();
|
||||
return TRUE;
|
||||
case WM_ENDSESSION:
|
||||
if (wp) { if (g_engine && !g_exited) sendEngine("quit"); }
|
||||
return 0;
|
||||
case WM_TIMER:
|
||||
if (wp == ID_RESTART_TIMER) { restartEngineNow(); return 0; }
|
||||
if (wp == ID_QUIT_TIMER) {
|
||||
DWORD code = 0;
|
||||
bool gone = !g_engine || (GetExitCodeProcess(g_engine, &code) && code != STILL_ACTIVE);
|
||||
|
|
@ -435,7 +500,7 @@ int WINAPI wWinMain(HINSTANCE hInst, HINSTANCE, PWSTR, int) {
|
|||
HANDLE once = CreateMutexW(nullptr, TRUE, L"Local\\IgneumMinerWindow");
|
||||
if (GetLastError() == ERROR_ALREADY_EXISTS) {
|
||||
HWND other = FindWindowW(L"IgneumMinerWindow", nullptr);
|
||||
if (other) { ShowWindow(other, SW_SHOW); SetForegroundWindow(other); }
|
||||
if (other) { PostMessageW(other, WM_ENGINE_RESTART, 0, 0); ShowWindow(other, SW_SHOW); SetForegroundWindow(other); }
|
||||
return 0;
|
||||
}
|
||||
CoInitializeEx(nullptr, COINIT_APARTMENTTHREADED);
|
||||
|
|
@ -473,7 +538,7 @@ int WINAPI wWinMain(HINSTANCE hInst, HINSTANCE, PWSTR, int) {
|
|||
TranslateMessage(&msg);
|
||||
DispatchMessageW(&msg);
|
||||
}
|
||||
if (g_engine) { CloseHandle(g_engine); }
|
||||
closeEngine();
|
||||
CloseHandle(once);
|
||||
CoUninitialize();
|
||||
return 0;
|
||||
|
|
|
|||
|
|
@ -85,6 +85,8 @@ Name: "{autodesktop}\Igneum Miner"; Filename: "{app}\igneum-app.exe"; Parameters
|
|||
; Started as the signed-in user, not as administrator (the data lands in that user's %LOCALAPPDATA%).
|
||||
Filename: "{app}\igneum-app.exe"; Parameters: "--launch"; Description: "Start Igneum Miner now"; Flags: postinstall nowait skipifsilent runasoriginaluser
|
||||
; The over-the-air updater (packaging/ota, src/ota.rs) runs this installer /VERYSILENT /IGNOTA=1 and the app must come back by itself.
|
||||
; /IGNOTA=2 (helpers from 0.3.21 on, MF-11): the helper starts the app itself, polls the new engine and restores the kept
|
||||
; exe set when nothing answers; this entry stays silent so the two never race for the single-instance window.
|
||||
Filename: "{app}\igneum-app.exe"; Parameters: "--launch"; Flags: nowait runasoriginaluser; Check: OtaRelaunch
|
||||
|
||||
[UninstallRun]
|
||||
|
|
@ -95,7 +97,8 @@ Filename: "powershell.exe"; Parameters: "-NoProfile -ExecutionPolicy Bypass -Fil
|
|||
Type: filesandordirs; Name: "{app}"
|
||||
|
||||
[Code]
|
||||
// /IGNOTA=1: the app's own updater started this install; relaunch the app when the files are in.
|
||||
// /IGNOTA=1: the app's own updater (before 0.3.21) started this install; relaunch the app when the files are in.
|
||||
// /IGNOTA=2: the 0.3.21 helper started it and relaunches the app itself (src/ota.rs WIN_HELPER): no relaunch here.
|
||||
function OtaRelaunch: Boolean;
|
||||
begin
|
||||
Result := ExpandConstant('{param:IGNOTA|0}') = '1';
|
||||
|
|
|
|||
Loading…
Reference in a new issue