Merge pull request 'fix: delete analytics logs older than 2 days and give vector a 512 MiB cap' (#9) from fix/log-retention-vector-512 into main

README.md

@@ -1,6 +1,6 @@
 # supaspin
 
-Version 1.3.1
+Version 1.3.2
 
 Create, list, show and delete throwaway [Supabase](https://supabase.com) instances on a
 single-node [microk8s](https://microk8s.io) server, from a terminal menu.
@@ -17,7 +17,7 @@ endangering them. Nothing in it is specific to any one app.
 | Postgres | `<LAN IP>:544N` (instance 1 → 5441, 2 → 5442, …), bound on the LAN address only |
 | Studio (admin UI) | `http://<LAN IP>:300N` (3001, 3002, …), not on the public hostname |
 | Disk | fixed-size ext4 image files, mounted at boot: **db 2 GB, storage 1 GB, every other volume 256 MB** |
-| Memory | limits add up to **9.5 GiB** per instance, requests to 1.875 GiB (table below) |
+| Memory | limits add up to **9 GiB** per instance, requests to 1.875 GiB (table below) |
 | Priority | pods run at priority -100 (`supaspin-low`); when the server runs short of memory they are evicted first and never evict anyone else |
 | Secrets | generated fresh per instance: JWT secret, anon and service keys, Postgres password, Studio password, and the internal keys |
 | Uploads | stored in the instance's own storage image (no MinIO); deleted with the instance |
@@ -93,7 +93,7 @@ Supabase Helm chart (0.8.0, Postgres 17.6.1.136) once into `/var/lib/supaspin/ch
 | Show instance (and suspend or resume it) | the `.env` (secrets hidden unless you ask), the tunnel command, pod status and recent events - and the one line that stops this instance or starts it again, worded for whichever state it is in |
 | Delete instance | asks you to type the number; removes everything (see below) |
 | Copy instance values into another .env file | puts an instance's `.env` values into your app's `.env`, on this server or on another computer (see below) |
-| Apply current memory caps to running instances | puts this version's memory limits on instances that already exist, replacing their pods one service at a time; databases, uploads and `.env` files are not touched |
+| Apply current memory caps and log retention to running instances | puts this version's memory limits and the log-retention job on instances that already exist, replacing their pods one service at a time; databases, uploads and `.env` files are not touched |
 | Mail settings | view and change the saved default mail settings |
 | Settings | change any first-run setting |
 
@@ -262,11 +262,11 @@ stores the new memory limits and leaves it asleep; they take effect when you res
 |---|---|---|---|
 | db | 512 MiB | 2048 MiB | 273 MiB |
 | analytics | 128 MiB | 1536 MiB | 694 MiB |
-| vector | 128 MiB | 1024 MiB | 290 MiB |
+| vector | 128 MiB | 512 MiB | 290 MiB |
 | kong | 128 MiB | 768 MiB | 269 MiB |
 | realtime | 128 MiB | 768 MiB | 300 MiB |
 | auth, functions, imgproxy, meta, rest, storage, studio | 128 MiB each | 512 MiB each | 203 MiB (studio) |
-| **Total** | **1.875 GiB** | **9.5 GiB** | |
+| **Total** | **1.875 GiB** | **9 GiB** | |
 
 The request is what the scheduler sets aside for a service and can never be given to anything else.
 The limit is the ceiling the kernel kills the container at. They do different jobs here.
@@ -278,13 +278,21 @@ in. The limits are roughly twice the highest use measured across three running i
 
 The earlier limits sat right on top of normal use and the kernel kept killing the two logging
 services. Analytics (Logflare, which stores the logs Studio shows) used 612, 694 and 662 MiB of a
-768 MiB ceiling and was killed 36 times on one instance. Vector, the log collector that reads the
+768 MiB ceiling and was killed 36 times on one instance. Vector, the log collector that ships the
-log files of every pod on the server and not only its own instance's, was killed 297 times on that
+instance's pod logs to analytics, was killed 297 times on that same instance, where a backlog it
-same instance, where a backlog it could not deliver kept growing in its memory. The two feed each
+could not deliver kept growing in its memory. The two feed each other: analytics dies, vector
-other: analytics dies, vector cannot deliver, vector fills up and dies, and on restart it replays
+cannot deliver, vector fills up and dies, and on restart it replays the backlog into analytics.
-the backlog into analytics. Realtime at 300 MiB of 384 and studio at 203 MiB of 256 were the next
+Realtime at 300 MiB of 384 and studio at 203 MiB of 256 were the next in line. Image resizing
-in line. Image resizing (imgproxy) reads the uploads from the storage volume, so it has no image of
+(imgproxy) reads the uploads from the storage volume, so it has no image of its own.
-its own.
+
+On 2026-10-06 instance 3 died of the same pipeline in a different place: Logflare stores the logs
+it receives in the `_supabase` database and never deletes them, so the 2 GB db image filled to
+100% over eight days, Postgres crashed and could not even start its recovery, analytics answered
+every delivery with an error, and vector was killed against its 1024 MiB ceiling every few
+minutes with the undeliverable backlog. The fix is log retention: every instance gets a pg_cron
+job (set up on create, upgrade and resume) that deletes analytics logs older than **2 days** every
+night. That bounds both the db image and vector's backlog, so vector is back to the 512 MiB
+ceiling the production supabase charts carry.
 
 Changing the numbers in this table only affects instances created afterwards. To put them on
 instances that already run, use `sudo supaspin upgrade`: it reads back the values Helm stored when

VERSION

@@ -1 +1 @@
-1.3.1
+1.3.2

supaspin

@@ -39,7 +39,7 @@ import tty
 
 import yaml
 
-VERSION = "1.3.1"
+VERSION = "1.3.2"
 CONFIG_PATH = os.environ.get("SUPASPIN_CONFIG", "/etc/supaspin/config.yaml")
 STATE_DIR = os.environ.get("SUPASPIN_STATE", "/var/lib/supaspin")
 CHART_REPO = "https://supabase-community.github.io/supabase-kubernetes"
@@ -72,14 +72,42 @@ DB_UID, DB_GID = 100, 101
 #   kong      260 / 269 / 265 MiB against 512
 #   studio 203, storage 187, rest 162 MiB against 256
 #   db        163 / 175 / 273 MiB against 2048 - left alone, it has room
+# On 2026-10-06 instance 3 filled its 2 GB database image with logs: analytics (Logflare) never
+# deletes what it stored, the db died of it, and vector's undelivered backlog was killed against
+# the 1024 MiB ceiling every few minutes. The fix is setup_log_retention: a pg_cron job deletes
+# analytics logs older than LOG_RETENTION in every instance, which also bounds vector's backlog,
+# so vector is back to the 512 MiB ceiling the production supabase charts carry.
 # Raising a limit costs nothing until a container actually asks for the memory; raising a request
 # would take the memory away from everything else on the node straight away.
 SERVICES = ["db", "analytics", "kong", "auth", "functions", "imgproxy", "meta",
             "realtime", "rest", "storage", "studio", "vector"]
 MEMORY = {"db": (512, 2048), "analytics": (128, 1536), "kong": (128, 768), "realtime": (128, 768),
-          "vector": (128, 1024)}
+          "vector": (128, 512)}
 DEFAULT_MEMORY = (128, 512)
 
+# Analytics logs: deleted from every instance's _supabase database when older than this. Logflare
+# stores logs in _analytics.log_events_* tables and never deletes them; without this job the
+# capped db image fills up and the database dies (instance 3, 2026-10-06).
+LOG_RETENTION = "2 days"
+LOG_RETENTION_JOB = "supaspin-log-retention"
+RETENTION_SQL = f"""
+create extension if not exists pg_cron;
+create or replace function _analytics.delete_old_logs() returns void language plpgsql as $fn$
+declare
+  t text;
+begin
+  for t in
+    select tablename from pg_tables
+    where schemaname = '_analytics' and tablename like 'log_events\\_%'
+  loop
+    execute format('delete from _analytics.%I where timestamp < now() - interval ''{LOG_RETENTION}''', t);
+  end loop;
+end $fn$;
+select _analytics.delete_old_logs();
+delete from cron.job where jobname = '{LOG_RETENTION_JOB}';
+select cron.schedule('{LOG_RETENTION_JOB}', '17 3 * * *', 'select _analytics.delete_old_logs();');
+""".strip()
+
 DEFAULT_MAIL = {"host": "smtp.example.com", "port": 587, "user": "", "password": "",
                 "admin_email": "admin@example.com", "sender_name": "Supabase"}
 
@@ -768,6 +796,19 @@ def wait_ready(n, log, timeout=1200):
     raise RuntimeError("pods did not become ready in time")
 
 
+def setup_log_retention(cfg, n, log):
+    """Give the instance a pg_cron job that deletes analytics logs older than LOG_RETENTION every
+    night. Logflare stores the logs it receives in the _supabase database and never deletes them;
+    without this job the capped db image fills up and the database dies (instance 3 on
+    2026-10-06). Runs the first cleanup right away, so an upgrade also frees a full disk. Runs
+    after create, upgrade and resume; scheduling under a fixed job name makes it idempotent."""
+    log(f"  log retention: deleting analytics logs older than {LOG_RETENTION}, every night")
+    kubectl("-n", namespace(n), "exec", f"{fullname(n)}-db-0", "-c", "supabase-db", "--",
+            "psql", "-U", "supabase_admin", "-d", "_supabase", "-v", "ON_ERROR_STOP=1",
+            "-c", RETENTION_SQL)
+    log("  log retention is set")
+
+
 def create_instance(cfg, mail, log, number=None):
     taken = existing_numbers(cfg)
     n = number if number is not None else lowest_free_number(cfg["slots"], taken)
@@ -796,6 +837,7 @@ def create_instance(cfg, mail, log, number=None):
         finally:
             os.remove(vfile)  # holds every secret
         wait_ready(n, log)
+        setup_log_retention(cfg, n, log)
         env_path = f"{inst_dir(cfg, n)}/.env"
         fd = os.open(env_path, os.O_WRONLY | os.O_CREAT | os.O_TRUNC, 0o600)
         with os.fdopen(fd, "w") as f:
@@ -891,6 +933,7 @@ def resume_instance(cfg, n, log):
     apply_values(cfg, n, set_replicas(values, 1))
     log("  the database answers first; the services that need it retry until it does")
     wait_ready(n, log)
+    setup_log_retention(cfg, n, log)
     log(f"  instance {n} is running again: https://{cfg['slots'].get(n, '?')}")
     return True
 
@@ -917,9 +960,10 @@ def upgrade_instance(cfg, n, log):
     if asleep:
         # A suspended instance stays suspended: the new caps are stored and take effect the moment
         # it is resumed. Waiting for its pods would wait for ever - by design it has none.
-        log(f"  instance {n} is suspended; the new caps apply when you resume it")
+        log(f"  instance {n} is suspended; the new caps and log retention apply when you resume it")
         return bool(changed)
     wait_ready(n, log)
+    setup_log_retention(cfg, n, log)
     log(f"  instance {n} is running with the current settings")
     return bool(changed)
 

tests/test_supaspin.py

@@ -90,14 +90,17 @@ class Values(unittest.TestCase):
 
     def test_memory_caps_add_up(self):
         req, lim = sp.total_memory_mib()
-        # db 2048, analytics 1536, kong 768, realtime 768, vector 1024, the other seven 512 each
+        # db 2048, analytics 1536, kong 768, realtime 768, vector 512, the other seven 512 each
-        self.assertEqual(lim, 2048 + 1536 + 768 + 768 + 1024 + 7 * 512)
+        self.assertEqual(lim, 2048 + 1536 + 768 + 768 + 512 + 7 * 512)
-        self.assertEqual(lim / 1024, 9.5)
+        self.assertEqual(lim / 1024, 9.0)
         self.assertEqual(req, 512 + 11 * 128)  # 1.875 GiB
         # Measured on three running instances on 2026-09-22; the ceiling has to stay clear of the
         # highest use, or the kernel kills the container in a loop the way it did on instance 1.
         self.assertGreaterEqual(sp.MEMORY["analytics"][1], 1536)  # used 694 MiB, died at 768
-        self.assertGreaterEqual(sp.MEMORY["vector"][1], 1024)     # used 290 MiB, died at 512
+        self.assertGreaterEqual(sp.MEMORY["vector"][1], 512)      # used 290 MiB; the nightly log
+        #  retention keeps its delivery backlog from growing past this the way it did on
+        #  2026-10-06, when instance 3's full db image killed analytics and vector filled with
+        #  the undeliverable backlog every few minutes against 1024 MiB.
         self.assertGreaterEqual(sp.MEMORY["realtime"][1], 768)    # used 300 MiB against 384
         self.assertGreaterEqual(sp.MEMORY["kong"][1], 768)        # used 269 MiB against 512
         self.assertGreaterEqual(sp.DEFAULT_MEMORY[1], 512)        # studio used 203 against 256
@@ -128,6 +131,15 @@ class Values(unittest.TestCase):
         self.assertEqual(set(sp.VOLUMES), {"db", "storage", "functions", "snippets", "deno", "pgsodium"})
         self.assertTrue(all(p[v]["storageClassName"] == "supaspin-local-sc" for v in sp.VOLUMES))
 
+    def test_log_retention_sql(self):
+        # Logflare stores logs in _analytics.log_events_* and never deletes them; the job deletes
+        # rows older than LOG_RETENTION and is scheduled under a fixed name so re-runs overwrite.
+        self.assertIn("schemaname = '_analytics'", sp.RETENTION_SQL)
+        self.assertIn("like 'log_events\\_%'", sp.RETENTION_SQL)
+        self.assertIn(f"interval ''{sp.LOG_RETENTION}''", sp.RETENTION_SQL)
+        self.assertIn(f"'{sp.LOG_RETENTION_JOB}'", sp.RETENTION_SQL)
+        self.assertIn("cron.schedule", sp.RETENTION_SQL)
+
     def test_imgproxy_reads_the_storage_volume(self):
         img = self.v["persistence"]["imgproxy"]
         self.assertEqual(img["existingClaim"], "supa2-supabase-storage")
@@ -975,6 +987,7 @@ class Upgrade(unittest.TestCase):
 
         with mock.patch.object(sp, "helm", helm), \
                 mock.patch.object(sp, "wait_ready", lambda n, log, **kw: None), \
+                mock.patch.object(sp, "setup_log_retention", lambda cfg, n, log: None), \
                 mock.patch.object(sp, "STATE_DIR", tempfile.mkdtemp()):
             yield
 
@@ -1147,6 +1160,8 @@ class SuspendResume(unittest.TestCase):
                                   lambda n, log, **kw: self.waited.append(("ready", n))), \
                 mock.patch.object(sp, "wait_gone",
                                   lambda n, log, **kw: self.waited.append(("gone", n))), \
+                mock.patch.object(sp, "setup_log_retention",
+                                  lambda cfg, n, log: self.waited.append(("retention", n))), \
                 mock.patch.object(sp, "STATE_DIR", tempfile.mkdtemp()):
             yield
 
@@ -1210,7 +1225,7 @@ class SuspendResume(unittest.TestCase):
         for svc in sp.SERVICES:
             self.assertEqual(self.written["deployment"][svc]["replicaCount"], 1)
         self.assertFalse(sp.is_suspended(self.written))
-        self.assertEqual(self.waited, [("ready", 3)])
+        self.assertEqual(self.waited, [("ready", 3), ("retention", 3)])
 
     def test_resuming_a_running_instance_changes_nothing(self):
         with self.fake_helm(self.RUNNING):
@@ -1237,7 +1252,7 @@ class SuspendResume(unittest.TestCase):
     def test_upgrade_still_waits_for_a_running_instance(self):
         with self.fake_helm(self.RUNNING):
             sp.upgrade_instance(CFG, 1, self.log.append)
-        self.assertEqual(self.waited, [("ready", 1)])
+        self.assertEqual(self.waited, [("ready", 1), ("retention", 1)])
 
 
 class InstanceState(unittest.TestCase):