fix(supervisor+ci): self-restart on deploy, CI creates date-based tags

supervisor: after git checkout + uv sync, os.execv replaces the running
process so new code takes effect immediately without a manual systemd
restart. systemd sees the same PID, so the unit stays "active".

ci: changed tag format from v{run_number} to v{YYYYMMDDHHMM}, matching
the supervisor's deploy tag convention. Sequential v<N> tags conflicted
with manual date-based tags causing an infinite redeploy loop.
No more manual tagging needed — CI tags automatically after green tests.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.gitea/workflows/ci.yaml

@@ -17,9 +17,9 @@ jobs:
       - run: uv run pytest web/tests/ -x -q -p no:faulthandler
       - run: uv run ruff check web/src/ web/tests/
 
-  # Creates v<N> tag after tests pass. The on-server supervisor polls for new
-  # tags every 60s and deploys automatically. No SSH keys or deploy credentials
-  # needed in CI — only the built-in github.token.
+  # Creates a v{YYYYMMDDHHMM} tag after tests pass on master.
+  # The on-server supervisor polls for new tags every 60s and deploys
+  # automatically. No SSH keys or deploy credentials needed in CI.
   tag:
     needs: [test]
     runs-on: ubuntu-latest
@@ -32,5 +32,6 @@ jobs:
         run: |
           git config user.name "CI"
           git config user.email "ci@noreply"
-          git tag "v${{ github.run_number }}"
-          git push origin "v${{ github.run_number }}"
+          TAG="v$(date -u +%Y%m%d%H%M)"
+          git tag "$TAG"
+          git push origin "$TAG"

src/padelnomics/supervisor.py

@@ -279,12 +279,18 @@ def web_code_changed() -> bool:
 
 
 def current_deployed_tag() -> str | None:
-    """Return the tag currently checked out, or None if not on a tag."""
+    """Return the highest-version tag pointing at HEAD, or None.
+
+    Uses the same sort order as latest_remote_tag() so that when multiple
+    tags point to the same commit (e.g. a date-based tag and a CI integer
+    tag), we always compare apples-to-apples.
+    """
     result = subprocess.run(
-        ["git", "describe", "--tags", "--exact-match", "HEAD"],
+        ["git", "tag", "--list", "--sort=-version:refname", "--points-at", "HEAD", "v*"],
         capture_output=True, text=True, timeout=10,
     )
-    return result.stdout.strip() or None
+    tags = result.stdout.strip().splitlines()
+    return tags[0] if tags else None
 
 
 def latest_remote_tag() -> str | None:
@@ -321,6 +327,10 @@ def git_pull_and_sync() -> None:
     run_shell(f"git checkout --detach {latest}")
     run_shell("sops --input-type dotenv --output-type dotenv -d .env.prod.sops > .env")
     run_shell("uv sync --all-packages")
+    # Re-exec so the new code is loaded. os.execv replaces this process in-place;
+    # systemd sees it as the same PID and does not restart the unit.
+    logger.info("Deploy complete — re-execing to load new code")
+    os.execv(sys.executable, sys.argv)
 
 
 # ---------------------------------------------------------------------------

transform/sqlmesh_padelnomics/models/staging/stg_population_geonames.sql

@@ -3,11 +3,7 @@
 -- Broad coverage (140K+ locations) enables Gemeinde-level market intelligence.
 -- One row per geoname_id (GeoNames stable numeric identifier).
 --
--- Supports two landing formats (UNION ALL during migration):
---   New: cities_global.jsonl.gz — one city per line, columns directly accessible
---   Old: cities_global.json.gz  — {"rows": [...]} blob (UNNEST required)
---
--- Source: data/landing/geonames/{year}/{month}/cities_global.{jsonl,json}.gz
+-- Source: data/landing/geonames/{year}/{month}/cities_global.jsonl.gz
 
 MODEL (
   name staging.stg_population_geonames,
@@ -16,21 +12,18 @@ MODEL (
   grain geoname_id
 );
 
-WITH
--- New format: one city per JSONL line
-jsonl_rows AS (
-  SELECT
+SELECT
   TRY_CAST(geoname_id  AS INTEGER)  AS geoname_id,
-    city_name,
-    country_code,
+  TRIM(city_name)                   AS city_name,
+  UPPER(country_code)               AS country_code,
   TRY_CAST(lat         AS DOUBLE)   AS lat,
   TRY_CAST(lon         AS DOUBLE)   AS lon,
-    admin1_code,
-    admin2_code,
+  NULLIF(TRIM(admin1_code), '')     AS admin1_code,
+  NULLIF(TRIM(admin2_code), '')     AS admin2_code,
   TRY_CAST(population  AS BIGINT)   AS population,
   TRY_CAST(ref_year    AS INTEGER)  AS ref_year,
   CURRENT_DATE                      AS extracted_date
-  FROM read_json(
+FROM read_json(
   @LANDING_DIR || '/geonames/*/*/cities_global.jsonl.gz',
   format = 'newline_delimited',
   columns = {
@@ -38,52 +31,10 @@ jsonl_rows AS (
     lat: 'DOUBLE', lon: 'DOUBLE', admin1_code: 'VARCHAR', admin2_code: 'VARCHAR',
     population: 'BIGINT', ref_year: 'INTEGER'
   }
-  )
-  WHERE geoname_id IS NOT NULL
-),
--- Old format: {"rows": [...]} blob — kept for transition
-blob_rows AS (
-  SELECT
-    TRY_CAST(row ->> 'geoname_id'  AS INTEGER)  AS geoname_id,
-    row ->> 'city_name'                          AS city_name,
-    row ->> 'country_code'                       AS country_code,
-    TRY_CAST(row ->> 'lat'         AS DOUBLE)    AS lat,
-    TRY_CAST(row ->> 'lon'         AS DOUBLE)    AS lon,
-    row ->> 'admin1_code'                        AS admin1_code,
-    row ->> 'admin2_code'                        AS admin2_code,
-    TRY_CAST(row ->> 'population'  AS BIGINT)    AS population,
-    TRY_CAST(row ->> 'ref_year'    AS INTEGER)   AS ref_year,
-    CURRENT_DATE                                 AS extracted_date
-  FROM (
-    SELECT UNNEST(rows) AS row
-    FROM read_json(
-      @LANDING_DIR || '/geonames/*/*/cities_global.json.gz',
-      auto_detect = true,
-      maximum_object_size = 40000000
-    )
-  )
-  WHERE (row ->> 'geoname_id') IS NOT NULL
-),
-all_rows AS (
-  SELECT * FROM jsonl_rows
-  UNION ALL
-  SELECT * FROM blob_rows
 )
-SELECT
-  geoname_id,
-  TRIM(city_name)                AS city_name,
-  UPPER(country_code)            AS country_code,
-  lat,
-  lon,
-  NULLIF(TRIM(admin1_code), '')  AS admin1_code,
-  NULLIF(TRIM(admin2_code), '')  AS admin2_code,
-  population,
-  ref_year,
-  extracted_date
-FROM all_rows
-WHERE population IS NOT NULL
+WHERE geoname_id IS NOT NULL
+  AND population IS NOT NULL
   AND population > 0
-  AND geoname_id IS NOT NULL
   AND city_name IS NOT NULL
   AND lat IS NOT NULL
   AND lon IS NOT NULL

web/scripts/init_landing_seeds.py

@@ -1,22 +1,19 @@
-"""Create minimal seed files for SQLMesh staging models that require landing data."""
+"""Create minimal seed files for SQLMesh staging models that require landing data.
+
+Seeds are empty JSONL gzip files — they satisfy DuckDB's file-not-found check
+while contributing zero rows to the staging models.
+"""
 import gzip
-import json
 from pathlib import Path
 
-seed = {
-    "date": "1970-01-01",
-    "captured_at_utc": "1970-01-01T00:00:00Z",
-    "venue_count": 0,
-    "venues_errored": 0,
-    "venues": [],
-}
-morning = Path("data/landing/playtomic/1970/01/availability_1970-01-01.json.gz")
-recheck = Path("data/landing/playtomic/1970/01/availability_1970-01-01_recheck_00.json.gz")
+# stg_playtomic_availability requires at least one morning and one recheck file
+morning = Path("data/landing/playtomic/1970/01/availability_1970-01-01.jsonl.gz")
+recheck = Path("data/landing/playtomic/1970/01/availability_1970-01-01_recheck_00.jsonl.gz")
 morning.parent.mkdir(parents=True, exist_ok=True)
 for p in [morning, recheck]:
     if not p.exists():
-        with gzip.open(p, "wt") as f:
-            json.dump(seed, f)
+        with gzip.open(p, "wb") as f:
+            pass  # empty JSONL — 0 rows, no error
         print("created", p)
     else:
         print("exists ", p)