fix(doccano-django): own skip-bootstrap replay mode

Signed-off-by: Akash Kumar <meakash7902@gmail.com>

Author: AkashKumar7902

doccano-django/Dockerfile

@@ -12,3 +12,10 @@
 # (PR #1889) and pipeline 3572 (PR #1964 minimal repro) where the
 # bug originally manifested.
 FROM doccano/doccano:backend
+
+USER root
+COPY doccano-entrypoint.sh /opt/bin/doccano-keploy-entrypoint.sh
+RUN chmod +x /opt/bin/doccano-keploy-entrypoint.sh
+USER doccano
+
+ENTRYPOINT ["/opt/bin/doccano-keploy-entrypoint.sh"]

doccano-django/README.md

@@ -30,10 +30,11 @@ that:
 ## What's in here
 
 * `Dockerfile` — thin wrapper around `doccano/doccano:backend` pinning
-  the upstream version this sample tracks. Future doccano releases
-  that change the bug-triggering shape are addressed by retagging
-  here, not by scattering version pins across the lane scripts in
-  `keploy/integrations` / `keploy/enterprise`.
+  the upstream version this sample tracks and installing the
+  sample entrypoint that honors `DOCCANO_SKIP_BOOTSTRAP=1`. Future
+  doccano releases that change the bug-triggering shape are addressed
+  by retagging here, not by scattering version pins across the lane
+  scripts in `keploy/integrations` / `keploy/enterprise`.
 * `docker-compose.yml` — the orchestration: postgres-13 alongside
   the doccano backend, on a fixed subnet so the lane scripts can
   rely on stable IPs across record/replay phases.
@@ -55,13 +56,20 @@ that:
 ```sh
 docker compose up -d
 ./flow.sh bootstrap
+
+docker compose down
+DOCCANO_SKIP_BOOTSTRAP=1 docker compose up -d
 ./flow.sh record-traffic
+
 docker compose down -v
 ```
 
 This is what the keploy/integrations and keploy/enterprise CI
 lanes wrap in `keploy record` / `keploy test` — the base compose
-is uninstrumented and runs unchanged inside those lanes.
+is uninstrumented and runs unchanged inside those lanes. The first
+launch runs doccano's migrations and creates the admin user; the
+second launch skips bootstrap and starts gunicorn directly against
+the populated database volume.
 
 ### Without keploy — measuring real Python line coverage
 
@@ -72,10 +80,14 @@ add `coverage.py` per-worker tracking:
 mkdir -p coverage
 docker compose -f docker-compose.yml -f docker-compose.coverage.yml up -d --build
 ./flow.sh bootstrap
+
+docker compose -f docker-compose.yml -f docker-compose.coverage.yml down
+DOCCANO_SKIP_BOOTSTRAP=1 docker compose -f docker-compose.yml -f docker-compose.coverage.yml up -d --build
 ./flow.sh record-traffic
+
 docker compose -f docker-compose.yml -f docker-compose.coverage.yml kill -s SIGTERM backend
 sleep 3
-docker compose -f docker-compose.yml -f docker-compose.coverage.yml up -d backend
+DOCCANO_SKIP_BOOTSTRAP=1 docker compose -f docker-compose.yml -f docker-compose.coverage.yml up -d backend
 ./flow.sh coverage
 docker compose -f docker-compose.yml -f docker-compose.coverage.yml down -v
 ```
@@ -91,17 +103,19 @@ ignore it and run the base compose, paying zero coverage cost.
 ```sh
 docker compose up -d
 ./flow.sh bootstrap
+docker compose down
 
 # In one shell:
-keploy record -c "docker compose up" --container-name doccano_backend \
+keploy record -c "DOCCANO_SKIP_BOOTSTRAP=1 docker compose up" --container-name doccano_backend \
   --proxy-port 18081 --dns-port 18082
 
 # In another shell:
 ./flow.sh record-traffic
 # SIGINT keploy when traffic returns
 
-keploy test -c "docker compose up" --containerName doccano_backend \
-  --apiTimeout 60 --delay 20 --proxy-port 18081 --dns-port 18082
+keploy test -c "DOCCANO_SKIP_BOOTSTRAP=1 docker compose up" --containerName doccano_backend \
+  --apiTimeout 60 --delay 20 --host 127.0.0.1 --port 18080 \
+  --proxy-port 18081 --dns-port 18082
 ```
 
 Expected outcome with the integrations fix in place: 0 failures,

doccano-django/doccano-entrypoint.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+if [ "${DOCCANO_SKIP_BOOTSTRAP:-0}" = "1" ]; then
+    echo "Making staticfiles"
+    if [ ! -d staticfiles ] || ! find staticfiles -mindepth 1 -print -quit | grep -q .; then
+        echo "Executing collectstatic"
+        python manage.py collectstatic --noinput
+    fi
+
+    echo "Starting django without bootstrap"
+    exec gunicorn \
+        "--bind=${HOST:-0.0.0.0}:${PORT:-8000}" \
+        "--workers=${WORKERS:-1}" \
+        --timeout=300 \
+        --capture-output \
+        --log-level info \
+        config.wsgi
+fi
+
+exec /opt/bin/prod-django.sh "$@"

doccano-django/docker-compose.yml

@@ -11,7 +11,9 @@
 #      DB state persists.
 #   2. DOCCANO_SKIP_BOOTSTRAP=1 → backend re-launches in
 #      gunicorn-only mode against the populated volume; recording /
-#      replay run against this incarnation.
+#      replay run against this incarnation. The sample defaults to one
+#      worker to keep Django's per-process ContentType cache
+#      deterministic across keploy replay.
 #
 # The split is what gives the lane a deterministic DB starting state
 # without paying the migration cost on every record/replay invocation.
@@ -39,6 +41,7 @@ services:
       DEBUG: "False"
       DJANGO_SETTINGS_MODULE: config.settings.production
       DOCCANO_SKIP_BOOTSTRAP: "${DOCCANO_SKIP_BOOTSTRAP:-0}"
+      WORKERS: "${DOCCANO_WORKERS:-1}"
     depends_on:
       postgres:
         condition: service_healthy

doccano-django/flow.sh

@@ -144,21 +144,21 @@ doccano_record_traffic() {
     # the recording captures nothing.
     doccano_wait_for_fixed_token 240 >/dev/null
 
-    # Worker-cache warmup. doccano runs 4 gunicorn workers; each
-    # worker keeps its own per-process Django ContentType cache and
-    # populates it lazily on the first polymorphic-resolver query
-    # that worker handles. Recording lanes that terminate with
-    # SIGINT (rather than waiting on a long --record-timer) need
-    # every worker's cache warmed before the explicit test traffic
-    # fires — otherwise cold workers fire their own
-    # django_content_type lookups at replay-time, find empty perTest
-    # cohorts, and the dependent endpoints return HTTP 500.
+    # Worker-cache warmup. The sample defaults to one gunicorn worker
+    # for deterministic keploy replay, but DOCCANO_WORKERS can raise
+    # that count for local experiments. Each worker keeps its own
+    # per-process Django ContentType cache and populates it lazily on
+    # the first polymorphic-resolver query that worker handles.
+    # Recording lanes that terminate with SIGINT (rather than waiting
+    # on a long --record-timer) need every worker's cache warmed before
+    # the explicit test traffic fires — otherwise cold workers fire
+    # their own django_content_type lookups at replay-time, find empty
+    # perTest cohorts, and the dependent endpoints return HTTP 500.
     #
-    # 4 workers × 4 requests = 16 calls is gunicorn-dispatch-jitter
-    # safe; /v1/me is the cheapest authenticated endpoint and
-    # exercises the same auth-token + ContentType chain as real
-    # test calls, so each iteration cleanly warms one worker's
-    # cache.
+    # The fixed 16 calls are gunicorn-dispatch-jitter safe for the
+    # previous 4-worker default; /v1/me is the cheapest authenticated
+    # endpoint and exercises the same auth-token + ContentType chain
+    # as real test calls.
     local warm_idx
     for warm_idx in $(seq 1 16); do
         curl -sS -H "$h_token" "$base/v1/me" >/dev/null 2>&1 || true