Merge branch 'main' into fix/automount-folder-on-session-creation

changes/589.fix

@@ -0,0 +1 @@
+Change the default values of `dependencies`, `callback_url` and `startup_command` to `None` instead of `undefined` (which is for template arguments).

changes/592.fix.md

@@ -0,0 +1 @@
+Temporarily rollback the default distributed lock implementation to `pg_advisory` from `etcd` until we identify and fix spurious lock timeout issues

config/sample.toml

@@ -85,7 +85,7 @@ hide-agents = false
 # "pg_advisory" uses PostgreSQL's session-level advisory lock.
 # "redlock" uses Redis-based distributed lock (Redlock) -- currently not supported.
 # "etcd" uses etcd-based distributed lock via etcetra.
-# distributed-lock = "etcd"
+# distributed-lock = "pg_advisory"
 
 # The Docker image name that is used for importing external Docker images.
 # You need to change this if your are at offline environments so that the manager

src/ai/backend/manager/api/session.py

@@ -631,13 +631,13 @@ async def _create(request: web.Request, params: dict[str, Any]) -> web.Response:
         t.Key('maxWaitSeconds', default=0) >> 'max_wait_seconds': t.Int[0:],
         tx.AliasedKey(['starts_at', 'startsAt'], default=None): t.Null | t.String,
         t.Key('reuseIfExists', default=True) >> 'reuse': t.ToBool,
-        t.Key('startupCommand', default=undefined) >> 'startup_command':
+        t.Key('startupCommand', default=None) >> 'startup_command':
             UndefChecker | t.Null | t.String,
         tx.AliasedKey(['bootstrap_script', 'bootstrapScript'], default=undefined):
             UndefChecker | t.Null | t.String,
-        t.Key('dependencies', default=undefined):
+        t.Key('dependencies', default=None):
             UndefChecker | t.Null | t.List(tx.UUID) | t.List(t.String),
-        tx.AliasedKey(['callback_url', 'callbackUrl', 'callbackURL'], default=undefined):
+        tx.AliasedKey(['callback_url', 'callbackUrl', 'callbackURL'], default=None):
             UndefChecker | t.Null | tx.URL,
         t.Key('owner_access_key', default=undefined): UndefChecker | t.Null | t.String,
     },

src/ai/backend/manager/config.py

@@ -251,7 +251,7 @@
         t.Key('ssl-cert', default=None): t.Null | tx.Path(type='file'),
         t.Key('ssl-privkey', default=None): t.Null | tx.Path(type='file'),
         t.Key('event-loop', default='asyncio'): t.Enum('asyncio', 'uvloop'),
-        t.Key('distributed-lock', default='etcd'):
+        t.Key('distributed-lock', default='pg_advisory'):
             t.Enum('filelock', 'pg_advisory', 'redlock', 'etcd'),
         t.Key('pid-file', default=os.devnull): tx.Path(
             type='file',