release: merge v1.8.1 to main

CHANGELOG.md

@@ -11,6 +11,19 @@ per the process in [`docs/releasing.md`](docs/releasing.md).
 
 ## [Unreleased]
 
+## [1.8.1] - 2026-07-18
+
+### Fixed
+
+- **One-click upgrade no longer shows a false "HTTP 502 — did not complete" on a
+  successful upgrade (#622).** The self-upgrade recreates the dashboard container
+  itself; while it restarts, caddy (the reverse proxy) stays up and answers
+  502/503/504 because the upstream is briefly gone. The result poller already rode
+  out a dropped connection but treated a gateway 5xx as terminal, so the modal
+  jumped to "failed" even though the upgrade landed. Gateway 502/503/504 are now
+  ridden out like a dropped connection; the durable control result is the real
+  outcome. A genuine backend 500 still fast-fails.
+
 ## [1.8.0] - 2026-07-17
 
 **Config UX round 2.** The Configuration view is regrouped around logical

VERSION

@@ -1 +1 @@
-1.8.0
+1.8.1

build/dashboard/mining_dashboard/web/static/configview.mjs

@@ -43,8 +43,9 @@ const UPGRADE_POLL_MAX = 450; // 15 minutes — an upgrade pulls a whole release
 // Poll /api/control/result until a terminal result lands; shared by the Configuration view and
 // the Upgrade button (#59). `skip` ignores an intermediate status under the same id (the
 // still-present "previewed" result while a commit runs; "running" while an upgrade runs). Both
-// flows recreate the dashboard container itself, so a fetch here can transiently fail (connection
-// refused mid-restart) — ride it out and keep polling until the result file answers.
+// flows recreate the dashboard container itself, so a fetch here can transiently fail — either a
+// dropped connection (proxy down) or a 502/503/504 (proxy up, upstream mid-restart, #622). Ride
+// both out and keep polling until the result file answers.
 async function pollResult(id, skip, max = POLL_MAX) {
   for (let i = 0; i < max; i++) {
     await new Promise((r) => setTimeout(r, POLL_MS));
@@ -55,6 +56,11 @@ async function pollResult(id, skip, max = POLL_MAX) {
       continue;
     }
     if (res.status === 202) continue;
+    // Both flows recreate the dashboard container itself; while it restarts, the reverse proxy
+    // (caddy) stays up and answers 502/503/504 — the upstream is briefly gone, not failed. Ride
+    // these out like a dropped connection (#59/#622); the durable control result is the real
+    // outcome and `max` is the backstop. A real backend 500 (upstream up, erroring) still throws.
+    if (res.status === 502 || res.status === 503 || res.status === 504) continue;
     if (!res.ok) throw new Error(`HTTP ${res.status}`);
     const out = await res.json();
     if (out.status === skip) continue;

build/dashboard/pyproject.toml

@@ -7,7 +7,7 @@ name = "mining-dashboard"
 # Keep in lockstep with the top-level VERSION file — the single source of truth for the stack version
 # (#44). A shell test (tests/stack/run.sh) fails if these drift; the dashboard *displays* the version
 # from VERSION (baked in as PITHEAD_VERSION, #58), so this is packaging metadata only.
-version = "1.8.0"
+version = "1.8.1"
 description = "Monitoring dashboard and XvB switching engine for Pithead"
 readme = "README.md"
 requires-python = ">=3.11"

build/dashboard/tests/frontend/configview.test.mjs

@@ -94,6 +94,22 @@ test("runUpgrade posts the seen version, skips 'running', rides out the restart,
   assert.equal(posted.headers["X-Pithead-Control"], "1"); // CSRF guard rides every mutation
 });
 
+test("runUpgrade rides out a 502/503/504 from the proxy (upstream mid-restart), not just a dropped connection (#622)", async () => {
+  let polls = 0;
+  const fetchStub = async (url) => {
+    if (url === "/api/control/upgrade") {
+      return { status: 202, ok: false, json: async () => ({ id: ID, status: "pending" }) };
+    }
+    polls++;
+    // caddy stays up and answers a gateway error while the dashboard upstream is recreated —
+    // the common self-upgrade state, and the one the old poller misclassified as terminal.
+    if (polls <= 3) return { status: 502, ok: false, json: async () => ({}) };
+    return okResult({ status: "upgraded", version: "v9.9.9" });
+  };
+  const out = await withFastPoll(fetchStub, () => runUpgrade("v9.9.9"));
+  assert.equal(out.status, "upgraded"); // rode out the 502s to the durable result, no throw
+});
+
 test("runUpgrade surfaces a host-side rejection as the outcome, not a throw", async () => {
   const fetchStub = async (url) =>
     url === "/api/control/upgrade"

pithead

@@ -859,6 +859,7 @@ doctor() {
     echo "Network:"
     check_stratum_exposure doctor
     check_stratum_listening
+    check_tor_running
     check_egress_firewall_installed
     check_tor_clearnet_egress
     # Clearnet initial sync (#183): a deliberate, privacy-relevant opt-in. Warn whenever it's on so
@@ -1763,6 +1764,24 @@ mining_stack_running() {
     container_is_running p2pool || container_is_running monerod || container_is_running xmrig-proxy
 }
 
+# doctor (#563): the tor container is unconditional (no compose profile gates it — it's the
+# Tor-first backbone in every deployment, remote-node mode included). So "revenue containers up
+# but tor down" is never a legitimate state: it means tor crashed or was stopped individually
+# while the stack kept mining, leaving clearnet dials no longer fail-closed and every off-box
+# connection (Healthchecks, Telegram, XvB, p2pool/Tari peers) without its Tor path. FAIL loudly so
+# doctor can't report all-clear on a silent privacy outage. A clean `down` (nothing running) is
+# fine — nothing to guard.
+check_tor_running() {
+    if container_is_running tor; then
+        dr_ok "Tor container is running — the privacy backbone is up."
+    elif mining_stack_running; then
+        dr_fail "The Tor container is DOWN while the mining stack is still running — the privacy backbone is dead: clearnet dials are no longer fail-closed and off-box connections (Healthchecks, Telegram, XvB, peers) have lost their Tor path. Restart it ('./pithead restart tor'; set tor.auto_heal:true to self-heal), or bring the stack down ('./pithead down')."
+    else
+        dr_info "Tor container isn't running — the stack is down (expected after './pithead down')."
+    fi
+    return 0
+}
+
 # doctor (#383): verify the #270 fail-closed egress rules are ACTUALLY installed while the stack
 # runs. `down` removes them, and a host reboot silently drops them while `restart: unless-stopped`
 # brings every container back — that reboot gap is the state this catches. Read-only: `sudo -n`
@@ -1776,11 +1795,10 @@ check_egress_firewall_installed() {
         return 0
     fi
     if ! container_is_running tor; then
-        if mining_stack_running; then
-            dr_fail "The Tor container is DOWN while the mining stack is still running — the privacy backbone is dead and clearnet dials are no longer fail-closed. Restart Tor ('./pithead restart tor'), or bring the stack down ('./pithead down')."
-        else
-            dr_info "Tor-egress firewall check skipped — the stack isn't running (rules are removed at 'down')."
-        fi
+        # tor down while the stack runs is caught by check_tor_running (a dedicated, loud verdict);
+        # here it's just an info-skip either way — a clean `down` removed the rules, and a tor-only
+        # outage is already being FAILed above.
+        dr_info "Tor-egress firewall check skipped — the tor container isn't running."
         return 0
     fi
     if ! command -v iptables >/dev/null 2>&1; then
@@ -1859,11 +1877,9 @@ check_dashboard_answers() {
 # must not fail cron health gates on a slow circuit.
 check_tor_clearnet_egress() {
     if ! container_is_running tor; then
-        if mining_stack_running; then
-            dr_fail "The Tor container is DOWN while the mining stack runs — every off-box connection (Healthchecks, Telegram, XvB, p2pool/Tari peers) has lost its Tor path. Restart Tor ('./pithead restart tor'; set tor.auto_heal:true to self-heal), or bring the stack down."
-        else
-            dr_info "Tor clearnet-egress check skipped — the tor container isn't running."
-        fi
+        # A tor-only outage is FAILed by check_tor_running; this test can't probe egress without a
+        # live tor either way, so it just info-skips.
+        dr_info "Tor clearnet-egress check skipped — the tor container isn't running."
         return 0
     fi
     if ! command -v curl >/dev/null 2>&1; then

tests/stack/run.sh

@@ -202,12 +202,17 @@ rm -f "$SANDBOX/.env"
 # Stack down (tor not running) -> info skip: rules are legitimately absent after 'down'.
 out="$(RUNNING_CONTAINERS="" PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_egress_firewall_installed 2>&1)"
 assert_contains "egress check: stack down -> info" "$out" "isn't running"
-# tor DOWN while the mining stack runs -> FAIL, not a silent skip: the privacy backbone is dead but
-# revenue containers are live (tor crashed / stopped individually). doctor must not report all-clear.
+# The tor-down-while-mining verdict lives in the dedicated check_tor_running (#563), so the egress
+# checks just info-skip when tor is down — they don't double-FAIL the same root cause.
 out="$(RUNNING_CONTAINERS="p2pool" PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_egress_firewall_installed 2>&1)"
-assert_contains "egress check: tor down + mining up -> FAIL" "$out" "Tor container is DOWN"
-out="$(RUNNING_CONTAINERS="p2pool" PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_tor_clearnet_egress 2>&1)"
-assert_contains "clearnet-egress check: tor down + mining up -> FAIL" "$out" "Tor container is DOWN"
+assert_contains "egress check: tor down -> info skip (dedicated check owns the verdict)" "$out" "isn't running"
+# check_tor_running: the loud, dedicated privacy-outage verdict.
+out="$(RUNNING_CONTAINERS="tor" PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_tor_running 2>&1)"
+assert_contains "tor-running check: tor up -> OK" "$out" "privacy backbone is up"
+out="$(RUNNING_CONTAINERS="" PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_tor_running 2>&1)"
+assert_contains "tor-running check: whole stack down -> info" "$out" "stack is down"
+out="$(RUNNING_CONTAINERS="p2pool" PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_tor_running 2>&1)"
+assert_contains "tor-running check: tor down + mining up -> FAIL" "$out" "Tor container is DOWN"
 # Running + tagged rules present -> OK.
 out="$(RUNNING_CONTAINERS="tor" IPT_TAGGED=1 PATH="$DRBIN:$PATH" run_sourced "$SANDBOX" check_egress_firewall_installed 2>&1)"
 assert_contains "egress check: rules installed -> OK" "$out" "installed"