fix: flakiness in railway deployment logs (#896)

* fix: handle flakiness in railway deployment logs by verifying actual deployment status

* fix: improve handling of railway deployment log-streaming errors to ensure accurate deployment status

* fix: enhance deployment status verification to address log-streaming glitches in railway

* fix: allow running of railway workflow for forked version

* Apply suggestions from code review

Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

---------

Co-authored-by: Rotimi Best <best@classroomio.com>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

.github/workflows/railway-preview.yml

@@ -22,6 +22,11 @@ name: Railway Preview Environment
 on:
   workflow_dispatch:
     inputs:
+      repo:
+        description: "Repo to deploy from (leave blank for classroomio/classroomio; set to e.g. Arun5768/classroomio for a fork)"
+        required: false
+        type: string
+        default: ""
       branch:
         description: "Branch (or PR head ref) to deploy into the preview env"
         required: true
@@ -61,6 +66,7 @@ jobs:
       - name: Checkout target branch
         uses: actions/checkout@v4
         with:
+          repository: ${{ inputs.repo != '' && inputs.repo || github.repository }}
           ref: ${{ inputs.branch }}
 
       - name: Install Railway CLI
@@ -244,10 +250,56 @@ jobs:
         if: ${{ inputs.action == 'deploy' }}
         # `--ci` (not `--detach`) blocks until each service's deploy/healthcheck succeeds,
         # so the URL printed later is actually live. Trade-off: builds run sequentially.
+        #
+        # `railway up --ci` can exit 1 on a log-streaming glitch (railwayapp/cli#696) even
+        # though the deploy succeeded. Only on that specific error do we poll the real
+        # deployment status instead of trusting the exit code — any other failure means no
+        # new deployment exists yet, so it fails immediately rather than risking a match
+        # against a stale, unrelated deployment.
         run: |
           for svc in $APP_SERVICES; do
             echo "::group::railway up $svc"
-            railway up --service "$svc" --ci
+            log_file=$(mktemp)
+            railway up --service "$svc" --ci 2>&1 | tee "$log_file"
+            railway_status=${PIPESTATUS[0]}
+            if [ "$railway_status" -ne 0 ]; then
+              if grep -q "Failed to stream build logs:" "$log_file"; then
+                deploy_id=$(grep -oP '(?<=[?&]id=)[0-9a-fA-F-]+' "$log_file" | head -n1)
+                if [ -z "$deploy_id" ]; then
+                  echo "::error::railway up failed for $svc and no deployment id could be recovered from its output — cannot verify actual status."
+                  rm -f "$log_file"
+                  exit 1
+                fi
+                echo "railway up reported failure for $svc due to the known log-streaming glitch; checking status of deployment $deploy_id..."
+                status=""
+                for attempt in $(seq 1 60); do
+                  status=$(railway deployment list --service "$svc" --json 2>/dev/null | jq -r --arg id "$deploy_id" '.[] | select(.id == $id) | .status // empty')
+                  case "$status" in
+                    SUCCESS)
+                      echo "Deployment $deploy_id for $svc actually succeeded (status: SUCCESS) — the CLI error was just a log-streaming glitch. Continuing."
+                      break
+                      ;;
+                    FAILED|CRASHED|REMOVED)
+                      echo "::error::$svc deployment $deploy_id did not succeed (status: $status)."
+                      rm -f "$log_file"
+                      exit 1
+                      ;;
+                  esac
+                  if [ "$attempt" -eq 60 ]; then
+                    echo "::error::$svc deployment $deploy_id never reached a terminal status after 60 attempts (last seen: ${status:-unknown})."
+                    rm -f "$log_file"
+                    exit 1
+                  fi
+                  echo "Deployment $deploy_id for $svc not finished yet (status: ${status:-unknown}, attempt $attempt/60) — retrying in 10s..."
+                  sleep 10
+                done
+              else
+                echo "::error::railway up failed for $svc for a reason other than the known log-streaming glitch — not polling stale deployment status."
+                rm -f "$log_file"
+                exit 1
+              fi
+            fi
+            rm -f "$log_file"
             echo "::endgroup::"
           done