feat(ifc file importer): add healthcheck endpoint (#5232)

* Serve path `/healthz`
* Include readiness and startup probes

packages/fileimport-service/src/nextGen/healthcheck.ts

@@ -14,7 +14,7 @@ export const startHealthCheckServer = (params: { logger: Logger }) => {
   })
 
   server.listen(9080, '0.0.0.0', () => {
-    logger.info('Server running, listening to 0.0.0.0 and path /healthz')
+    logger.info('Healthcheck server is running. Serving at 0.0.0.0:9080/healthz')
   })
 
   return server

packages/ifc-import-service/main.py

@@ -1,6 +1,8 @@
 import asyncio
 import logging
 import sys
+import threading
+from http.server import BaseHTTPRequestHandler, HTTPServer
 
 import structlog
 from structlog_to_seq import CelfProcessor
@@ -31,9 +33,25 @@ def configure_logger() -> structlog.stdlib.BoundLogger:
     return logger
 
 
+class HealthcheckHTTPRequestHandler(BaseHTTPRequestHandler):
+    def do_GET(self):  # noqa: N802
+        match self.path:
+            case "/healthz":
+                self.send_response(200)
+                self.send_header("Content-type", "application/json")
+                self.end_headers()
+                self.wfile.write(b'{"status": "OK"}')
+            case _:
+                self.send_response(404)
+                self.end_headers()
+
+
 async def main():
     logger = configure_logger()
     task = asyncio.create_task(job_processor(logger))
+    httpd = HTTPServer(("0.0.0.0", 9080), HealthcheckHTTPRequestHandler)
+    thread = threading.Thread(target=httpd.serve_forever, daemon=True)
+    thread.start()
 
     # we do not need any sort of signal handling logic,
     # cause if the context of the job transaction exits,

packages/ifc-import-service/src/ifc_importer/job_processor.py

@@ -69,11 +69,13 @@ async def job_handler(
 
 async def job_processor(logger: structlog.stdlib.BoundLogger):
     parser = "speckle_ifc"
+    logger = logger.bind(parser=parser)
     connection = await setup_connection()
+    logger.info("job processor started")
     while True:
         job = await get_next_job(connection)
         if not job:
-            logger.info("no job found")
+            logger.debug("no job found")
             await asyncio.sleep(IDLE_TIMEOUT)
             continue
 

utils/helm/speckle-server/templates/fileimport_service/deployment.yml

@@ -37,9 +37,42 @@ spec:
             containerPort: 9093
             protocol: TCP
 
+        startupProbe:
+          periodSeconds: 10
+          failureThreshold: 18 # 10 * 18 = 180s to startup
+          timeoutSeconds: 3
+          {{- if .Values.featureFlags.nextGenFileImporterEnabled }}
+          httpGet:
+            path: /healthz
+            port: 9080
+          {{- else }}
+          exec:
+            command:
+              - /usr/bin/node
+              - -e
+              - "process.exit((Date.now() - require('fs').readFileSync('/tmp/last_successful_query', 'utf8') > 25 * 60 * 1000) ? 1 : 0)"
+          {{- end }}
+
         livenessProbe:
           initialDelaySeconds: 60
           periodSeconds: 60
+          timeoutSeconds: 3
+          {{- if .Values.featureFlags.nextGenFileImporterEnabled }}
+          httpGet:
+            path: /healthz
+            port: 9080
+          {{- else }}
+          exec:
+            command:
+              - /usr/bin/node
+              - -e
+              - "process.exit((Date.now() - require('fs').readFileSync('/tmp/last_successful_query', 'utf8') > 25 * 60 * 1000) ? 1 : 0)"
+          {{- end }}
+
+        readinessProbe:
+          initialDelaySeconds: 60
+          periodSeconds: 10
+          timeoutSeconds: 3
           {{- if .Values.featureFlags.nextGenFileImporterEnabled }}
           httpGet:
             path: /healthz

utils/helm/speckle-server/templates/ifc_import_service/deployment.yml

@@ -31,13 +31,29 @@ spec:
             containerPort: 9093
             protocol: TCP
 
-        # TODO: Enable health checks
-        # livenessProbe:
-        #   initialDelaySeconds: 60
-        #   periodSeconds: 60
-        #   httpGet:
-        #     path: /healthz
-        #     port: 9080
+        startupProbe:
+          periodSeconds: 10
+          failureThreshold: 18 # 10 * 18 = 180s to startup
+          timeoutSeconds: 3
+          httpGet:
+            path: /healthz
+            port: 9080
+
+        livenessProbe:
+          initialDelaySeconds: 60
+          periodSeconds: 60
+          timeoutSeconds: 3
+          httpGet:
+            path: /healthz
+            port: 9080
+
+        readinessProbe:
+          initialDelaySeconds: 60
+          periodSeconds: 10
+          timeoutSeconds: 3
+          httpGet:
+            path: /healthz
+            port: 9080
 
         resources:
           {{- with .Values.ifc_import_service.requests }}