From 2469b391ec7ff7a7c6e8389436326eadffb4fab4 Mon Sep 17 00:00:00 2001
From: Nilanjan Chatterjee <nilanjan.9325@gmail.com>
Date: Sun, 11 Oct 2026 02:30:19 -0500
Subject: [PATCH] fix(dashboard): scale worker table warning colors with
 worker-ttl

The worker table marked a worker red once its last heartbeat was more
than 60s old, regardless of distributed.scheduler.worker-ttl. With the
default 5 minute TTL that flags healthy-but-busy workers (e.g. ones
holding the GIL) long before the scheduler would act, and users had no
way to tune it.

Color rows orange after 25% and red after 50% of the effective TTL,
computed the same way as Scheduler.check_worker_ttl. When worker-ttl is
disabled, keep the old 60s red threshold.

Closes #8749

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 distributed/http/scheduler/info.py            | 36 +++++++++++++++-
 .../scheduler/tests/test_scheduler_http.py    | 43 +++++++++++++++++++
 distributed/http/templates/worker-table.html  |  3 +-
 3 files changed, 80 insertions(+), 2 deletions(-)

diff --git a/distributed/http/scheduler/info.py b/distributed/http/scheduler/info.py
index 78ad7f89e89..94cfacd4300 100644
--- a/distributed/http/scheduler/info.py
+++ b/distributed/http/scheduler/info.py
@@ -22,10 +22,44 @@
 
 if TYPE_CHECKING:
     from distributed import Scheduler
+    from distributed.scheduler import WorkerState
+
+#: Row colors in the worker table for workers that have not sent a heartbeat for a
+#: while, as (fraction of the worker TTL, color) in decreasing order of fraction.
+WORKER_TTL_COLORS = [(0.5, "#ffcdd2"), (0.25, "#ffe0b2")]
+#: Threshold in seconds used when ``distributed.scheduler.worker-ttl`` is disabled
+WORKER_NO_TTL_THRESHOLD = 60
+
+
+def worker_row_color(scheduler: Scheduler, ws: WorkerState) -> str | None:
+    """Background color for a worker in the worker table, or None.
+
+    The thresholds scale with the timeout after which the scheduler removes a
+    worker that stopped sending heartbeats (``distributed.scheduler.worker-ttl``),
+    so that the table flags a worker before the scheduler gives up on it.
+    """
+    from distributed.scheduler import heartbeat_interval
+
+    since = time() - ws.last_seen
+    if not scheduler.worker_ttl:
+        return WORKER_TTL_COLORS[0][1] if since > WORKER_NO_TTL_THRESHOLD else None
+    # Same effective TTL as Scheduler.check_worker_ttl
+    ttl = max(scheduler.worker_ttl, 10 * heartbeat_interval(len(scheduler.workers)))
+    for fraction, color in WORKER_TTL_COLORS:
+        if since > fraction * ttl:
+            return color
+    return None
+
 
 ns = {
     func.__name__: func
-    for func in [format_bytes, format_time, datetime.fromtimestamp, time]
+    for func in [
+        format_bytes,
+        format_time,
+        datetime.fromtimestamp,
+        time,
+        worker_row_color,
+    ]
 }
 
 rel_path_statics = {"rel_path_statics": "../../.."}
diff --git a/distributed/http/scheduler/tests/test_scheduler_http.py b/distributed/http/scheduler/tests/test_scheduler_http.py
index b7aac70121a..02fdf693d3c 100644
--- a/distributed/http/scheduler/tests/test_scheduler_http.py
+++ b/distributed/http/scheduler/tests/test_scheduler_http.py
@@ -15,6 +15,7 @@
 from distributed import Event, Lock, Scheduler
 from distributed.client import Client, wait
 from distributed.core import Status
+from distributed.metrics import time
 from distributed.utils import is_valid_xml, url_escape
 from distributed.utils_test import (
     async_poll_for,
@@ -67,6 +68,48 @@ async def test_connect(c, s, a, b):
                 assert not re.search("href=./", body)  # no absolute links
 
 
+async def _worker_table_colors(s):
+    http_client = AsyncHTTPClient()
+    response = await http_client.fetch(
+        f"http://localhost:{s.http_server.port}/info/main/workers.html"
+    )
+    body = response.body.decode()
+    colors = {}
+    for row in re.findall(r"<tr[^>]*>\s*<td><a href=[^>]*>[^<]*</a>", body):
+        addr = re.search(r">([^<]*)</a>$", row).group(1)
+        color = re.search(r"background-color:(#[0-9a-f]+)", row)
+        colors[addr] = color.group(1) if color else None
+    return colors
+
+
+@gen_cluster(
+    nthreads=[("", 1)] * 3,
+    worker_kwargs={"heartbeat_interval": "100s"},
+    config={"distributed.scheduler.worker-ttl": "100s"},
+)
+async def test_worker_table_color_scales_with_worker_ttl(s, a, b, c):
+    s.workers[b.address].last_seen = time() - 30
+    s.workers[c.address].last_seen = time() - 70
+    colors = await _worker_table_colors(s)
+    assert colors == {
+        a.address: None,
+        b.address: "#ffe0b2",
+        c.address: "#ffcdd2",
+    }
+
+
+@gen_cluster(
+    nthreads=[("", 1)] * 2,
+    worker_kwargs={"heartbeat_interval": "100s"},
+    config={"distributed.scheduler.worker-ttl": None},
+)
+async def test_worker_table_color_without_worker_ttl(s, a, b):
+    s.workers[a.address].last_seen = time() - 30
+    s.workers[b.address].last_seen = time() - 70
+    colors = await _worker_table_colors(s)
+    assert colors == {a.address: None, b.address: "#ffcdd2"}
+
+
 @gen_cluster(client=True, nthreads=[])
 async def test_worker_404(c, s):
     http_client = AsyncHTTPClient()
diff --git a/distributed/http/templates/worker-table.html b/distributed/http/templates/worker-table.html
index 925473f7b2d..70adde25a05 100644
--- a/distributed/http/templates/worker-table.html
+++ b/distributed/http/templates/worker-table.html
@@ -20,7 +20,8 @@
     </thead>
     <tbody>
         {% for ws in worker_list %}
-        <tr {{ "style=background-color:#ffcdd2" if time() - ws.last_seen> 60 else ""}}>
+        {% set row_color = worker_row_color(scheduler, ws) %}
+        <tr {% if row_color %}style="background-color:{{ row_color }}"{% end %}>
             <td><a href="../worker/{{ url_escape(ws.address) }}.html">{{ws.address}}</a></td>
             <td> {{ ws.name if ws.name is not None else "" }} </td>
             <td> {{ ws.nthreads }} </td>