Use scaleset name as runner name prefix

charms/garm/src/scaleset_reconciler.py

@@ -84,6 +84,13 @@
 # GitHub's own limit on the System label GARM registers the scale set under.
 MAX_SCALESET_NAME_LENGTH = 64
 
+# GARM names each runner `<runner_prefix>-<id>` and never validates the prefix length.
+# The name becomes the OpenStack server name and so the runner's hostname, bounded to a
+# 63-character DNS label. The id is the base62 form of a shortid, or of a UUID when that
+# fails — at most 22 characters — which leaves 63 - 1 - 22 for the prefix.
+MAX_RUNNER_PREFIX_LENGTH = 40
+DEFAULT_RUNNER_PREFIX = "garm"
+
 # Past this, a runner a retired scaleset still reports cannot be doing a legitimate
 # job: these are self-hosted runners, bound by MAX_JOB_RUNTIME, not the six-hour cap
 # GitHub enforces on the runners it hosts itself. The margin absorbs reconcile
@@ -1047,6 +1054,7 @@ def _to_create_params(spec: ScalesetSpec, scaleset_name: str) -> CreateScaleSetP
                 "enable_shell": spec.enable_shell,
                 "extra_specs": _effective_extra_specs(spec) or None,
                 "template_id": spec.template_id,
+                "runner_prefix": runner_prefix(spec.name),
             }
         )
 
@@ -1103,6 +1111,7 @@ def _maybe_update(self, observed: ScaleSet, spec: ScalesetSpec, template_id: int
             enable_shell=spec.enable_shell,
             extra_specs=extra_specs,
             template_id=spec.template_id,
+            runner_prefix=runner_prefix(spec.name),
         )
         # Send template_id when the scaleset has, or had, a custom template — a 0
         # value detaches it (reverts to the default); omit it otherwise so an
@@ -1135,6 +1144,7 @@ def _needs_update(observed: ScaleSet, spec: ScalesetSpec, template_id: int) -> b
             != bool(desired_extra.get("disable_updates"))
             or observed_extra.get("image_visibility") != desired_extra.get("image_visibility")
             or (observed.template_id or 0) != template_id
+            or observed.runner_prefix != runner_prefix(spec.name)
         )
 
 
@@ -1171,6 +1181,22 @@ def _name_base(logical_name: str) -> str:
     return f"{logical_name[: limit - LABEL_HASH_LENGTH - 1]}-{digest}"
 
 
+def runner_prefix(logical_name: str) -> str:
+    """Return the prefix GARM names a scaleset's runners with.
+
+    Args:
+        logical_name: The scaleset name the operator configured.
+
+    Returns:
+        The name reduced to a hostname-safe form — lowercase alphanumerics joined by
+        single hyphens — truncated to ``MAX_RUNNER_PREFIX_LENGTH``. GARM's own default
+        prefix when nothing usable remains. Truncation needs no disambiguating hash:
+        GARM appends a random id to every runner name.
+    """
+    prefix = re.sub(r"[^a-z0-9]+", "-", logical_name.lower()).strip("-")
+    return prefix[:MAX_RUNNER_PREFIX_LENGTH].rstrip("-") or DEFAULT_RUNNER_PREFIX
+
+
 def _is_family_member(observed_name: str, logical_name: str) -> bool:
     """Return whether a live scaleset is a generation of *logical_name*.
 

charms/garm/tests/unit/test_scaleset_reconciler.py

@@ -15,12 +15,14 @@
 from runner_template import build_template_data
 from scaleset_reconciler import (
     DRAIN_DEADLINE,
+    MAX_RUNNER_PREFIX_LENGTH,
     MAX_SCALESET_NAME_LENGTH,
     Handover,
     ScalesetProgress,
     ScalesetReconciler,
     ScalesetSpec,
     _effective_extra_specs,
+    runner_prefix,
     target_scaleset_name,
 )
 
@@ -51,6 +53,7 @@ def __init__(
         template_id=None,
         enabled=True,
         updated_at=None,
+        runner_prefix=None,
     ):
         self.name = name
         self.id = sid
@@ -72,6 +75,7 @@ def __init__(
         # GARM stamps this on every write, so for a retired scaleset it is the
         # moment it was disabled — which is what the drain deadline measures from.
         self.updated_at = updated_at or datetime.now(timezone.utc)
+        self.runner_prefix = runner_prefix
 
 
 class _FakeJob:
@@ -151,6 +155,7 @@ def __init__(
                 template_id=ss.get("template_id", None),
                 enabled=ss.get("enabled", True),
                 updated_at=ss.get("updated_at", None),
+                runner_prefix=ss.get("runner_prefix", None),
             )
             for ss in (scalesets or [])
         ]
@@ -357,6 +362,7 @@ def _existing_scaleset(**overrides):
         extra_specs={},
         tags=[],
         enabled=True,
+        runner_prefix="my-scaleset",
     )
     base.update(overrides)
     return base
@@ -415,6 +421,65 @@ def test_no_update_when_scaleset_unchanged():
     assert client.deleted == []
 
 
+def test_create_names_runners_after_scaleset():
+    """
+    arrange: FakeGarmClient with the provider registered and no existing scalesets.
+    act: Reconcile a desired spec.
+    assert: The scaleset is created with its logical name as runner prefix, so runners
+        are named after the scaleset rather than GARM's default ``garm-<id>``.
+    """
+    client = FakeGarmClient(providers=["openstack-demo"], scalesets=[])
+
+    _reconcile(client, [_spec(name="my-scaleset")])
+
+    assert len(client.created) == 1
+    _, _, params = client.created[0]
+    assert params.runner_prefix == "my-scaleset"
+
+
+def test_update_renames_runners_of_existing_scaleset():
+    """
+    arrange: An existing scaleset still on GARM's default runner prefix.
+    act: Reconcile the matching desired spec.
+    assert: It is updated in place to the scaleset-derived prefix, so scalesets created
+        before this change also name new runners after themselves without a recreate.
+    """
+    client = FakeGarmClient(
+        providers=["openstack-demo"],
+        scalesets=[_existing_scaleset(runner_prefix="garm")],
+    )
+
+    _reconcile(client, [_spec()])
+
+    assert client.created == []
+    assert len(client.updated) == 1
+    _, params = client.updated[0]
+    assert params.runner_prefix == "my-scaleset"
+
+
+@pytest.mark.parametrize(
+    "logical_name, expected",
+    [
+        ("my-scaleset", "my-scaleset"),
+        ("My_Scale.Set", "my-scale-set"),
+        ("--x86__jammy--", "x86-jammy"),
+        ("a" * (MAX_RUNNER_PREFIX_LENGTH + 10), "a" * MAX_RUNNER_PREFIX_LENGTH),
+        ("a" * (MAX_RUNNER_PREFIX_LENGTH - 1) + "-b", "a" * (MAX_RUNNER_PREFIX_LENGTH - 1)),
+        ("___", "garm"),
+    ],
+    ids=["kept", "sanitised", "trimmed", "truncated", "no-trailing-hyphen", "empty-fallback"],
+)
+def test_runner_prefix(logical_name, expected):
+    """
+    arrange: A configured scaleset name.
+    act: Derive the runner prefix from it.
+    assert: The prefix is a lowercase hostname-safe form of the name, at most
+        MAX_RUNNER_PREFIX_LENGTH long so GARM's `<prefix>-<id>` runner name still fits a
+        63-character hostname, falling back to GARM's default when nothing usable is left.
+    """
+    assert runner_prefix(logical_name) == expected
+
+
 def test_update_when_scaleset_disabled():
     """
     arrange: FakeGarmClient with one existing disabled scaleset that otherwise matches.

docs/changelog.md

@@ -8,6 +8,10 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 Each revision is versioned by the date of the revision.
 
+## 2026-10-16
+
+- `garm`: name runners after their scale set (for example `my-scaleset-dvnyp5dm3sm`) instead of GARM's default `garm-<id>`. The prefix is the scale set name in lowercase, with every run of characters other than letters and digits replaced by a hyphen, and truncated to 40 characters so the runner name still fits a 63-character hostname. Existing scale sets are updated in place; runners created before the update keep their names.
+
 ## 2026-10-15
 
 - `garm`: bump the GARM workload so that it starts on PostgreSQL databases created by older GARM versions. On such databases, the previous workload failed a startup migration with `operator does not exist: text = uuid` and GARM did not start; GARM now converts the affected columns to the expected types on startup.