Real-Time Kong Interview Questions: Operations (2027)
40 real-time Kong interview questions on running Kong in production: health checks, kong drain, migrations, Prometheus metrics, decK rollback, incidents.
Real-time Kong interview questions test whether you can run Kong Gateway in production day to day: check node health, read the error log, drain a node before maintenance, upgrade with database migrations, watch metrics from the Prometheus plugin, and recover from bad configuration pushes. Interviewers want commands and checks, not definitions. This page has 40 operations questions with answers, grouped by level, plus live incident scenarios.
Commands and defaults were checked against Kong's official documentation at developer.konghq.com in November 2026. For the full concept list, start with the main guide, Kong API Gateway interview questions. This page covers the operator's side: what you do when Kong is already serving traffic.
Key takeaways
- Default listeners: proxy on 8000 (HTTP) and 8443 (HTTPS), Admin API on 127.0.0.1:8001 and 127.0.0.1:8444, hybrid mode control plane cluster listener on 8005.
- The Kong CLI covers daily operations:
kong check,kong health,kong reload,kong quit,kong drainand thekong migrationsfamily. kong drainmakes the/status/readyendpoint return 503, so load balancers and Kubernetes stop sending new traffic before shutdown.- Upgrades with a database follow
kong migrations up, roll the nodes, thenkong migrations finish. - The Prometheus plugin exposes metrics on the Status API (preferred) or the Admin API. Status code, latency, bandwidth and upstream health metrics are off until you enable them.
- decK lets you diff configuration before you sync it, which is the main guard against bad pushes.
How this page fits with the other Kong guides
| Page | Focus |
|---|---|
| Kong API Gateway interview questions | Main guide: entities, plugins, deployment modes |
| 75+ Kong questions: hands-on config | Admin API calls, plugin config, Ingress Controller YAML |
| Scenario-based Kong questions | Error codes such as 404 no route matched, 502 and 503 |
| Advanced Kong questions | Architecture, custom plugins, multi-region |
| This page | Running Kong: health, logs, metrics, maintenance, upgrades, rollbacks |
Kong operations questions for freshers
1. What ports does a Kong Gateway node listen on by default?
The configuration reference lists proxy_listen as 0.0.0.0:8000 and 0.0.0.0:8443 (with SSL and HTTP/2), and admin_listen as 127.0.0.1:8001 and 127.0.0.1:8444. In hybrid mode the control plane uses cluster_listen on port 8005. Kong Manager, where used, defaults to 8002 and 8445.
2. Why does the Admin API bind to 127.0.0.1 by default?
Because anyone who can reach the Admin API can change routing and security plugins. Keeping it on localhost forces you to decide deliberately how to expose it, for example behind Kong itself with authentication or on a private network only.
3. How do you check that a Kong node is healthy?
Run kong health on the node, which checks that the necessary services are running. Over HTTP, the Status API's /status endpoint reports server and database state, and /status/ready tells you whether the node is ready to proxy.
4. What does kong check do?
It validates a Kong configuration file (kong.conf) and reports errors without starting the node. Run it before every restart that follows a config change.
5. Where are Kong's logs?
By default in the logs directory under the Kong prefix: error.log for proxy errors and node messages, and access.log for proxied requests. In containers these are usually sent to stdout and stderr. The default log_level is notice.
6. What is the difference between kong reload, kong restart and kong quit?
kong reload sends SIGHUP to Nginx so new workers start with the new configuration while old ones finish requests. kong restart stops and starts the node. kong quit shuts down gracefully, forcing a stop if a timeout is reached.
7. What does a 404 "no Route matched with those values" mean in production?
Kong received the request but no route matched its host, path, method or headers. It is usually a configuration gap or a client calling the wrong host, not a backend failure.
8. What do 502 and 504 responses from Kong usually point to?
Problems between Kong and the upstream. A 502 often means Kong could not connect or got an invalid response; a 504 means the upstream did not answer within the configured timeouts. Check the error log entry for the request and the service's connect_timeout, read_timeout and write_timeout.
9. How do you see the Kong version on a node?
Run kong version. Add --all to include dependency versions. The Admin API root endpoint also returns the version.
10. What is decK used for in operations?
Exporting the current configuration, diffing a declarative file against what is running, and syncing changes. Teams store the file in Git and run deck gateway diff in CI before deck gateway sync.
Kong operations questions for 2 to 5 years of experience
11. How do you take a Kong node out of service without dropping requests?
Run kong drain so /status/ready returns 503. Load balancers and Kubernetes readiness probes that watch that endpoint stop sending new traffic. After in-flight requests finish, stop the node with kong quit.
12. Which endpoint should a Kubernetes readiness probe use?
/status/ready on the Status API port. It returns 200 only when the node is ready to proxy traffic, which also covers DB-less and data plane nodes that have not yet loaded configuration.
13. Where is the Status API exposed?
On the address set by status_listen. The current configuration reference lists the default as 127.0.0.1:8007. It is read-only, which makes it the right place for health checks and metrics scraping instead of the Admin API.
14. How do you collect metrics from Kong?
Enable the Prometheus plugin (globally or per service) and scrape /metrics. Kong's docs prefer the Status API for this; the Admin API is a fallback and does not work for scraping when Admin API RBAC is enabled.
curl -X POST http://localhost:8001/plugins \
--data name=prometheus \
--data config.status_code_metrics=true \
--data config.latency_metrics=true \
--data config.upstream_health_metrics=true
15. Which Kong metrics would you put on an on-call dashboard?
Request rate and status codes (kong_http_requests_total), latency split into Kong, upstream and total (kong_kong_latency_ms, kong_upstream_latency_ms, kong_request_latency_ms), upstream target health (kong_upstream_target_health), Nginx connections, Lua shared dictionary memory, and database reachability (kong_datastore_reachable).
16. How do you tell whether slowness is Kong or the backend?
Compare kong_kong_latency_ms (time spent in Kong and plugins) with kong_upstream_latency_ms (time waiting for the upstream). Kong also adds X-Kong-Proxy-Latency and X-Kong-Upstream-Latency response headers by default, which help when checking a single request.
17. How do you raise the log level temporarily without a restart?
Use the Admin API's debug endpoints (/debug/node/log-level) or the kong debug CLI command to change the log level on a running node, then set it back. Debug logs are verbose and can contain request data, so keep the window short.
18. How do you upgrade a Kong cluster that uses PostgreSQL?
Back up the database, read the upgrade notes for the target version, install the new version on one node, and run kong migrations up. Roll the remaining nodes to the new version, then run kong migrations finish once all old nodes are gone. kong migrations list shows the state at any point.
19. Why are migrations split into up and finish?
up applies changes that old and new nodes can both work with, so the cluster can run mixed versions during the rollout. finish applies the remaining changes that only the new version supports. Running finish too early can break nodes still on the old version.
20. How do you upgrade nodes in hybrid mode?
Upgrade the control plane first, then the data planes. Kong documents version compatibility rules between control and data planes, and the Prometheus plugin exposes kong_data_plane_version_compatible so you can see incompatible nodes.
21. How do you confirm a data plane has the latest configuration?
Compare kong_data_plane_config_hash across data planes and with the control plane, and check kong_data_plane_last_seen. The control plane's /clustering/data-planes Admin API endpoint lists connected data planes with their config hash and last seen time.
22. How do you monitor hybrid mode certificate expiry?
Alert on kong_data_plane_cluster_cert_expiry_timestamp. If the cluster certificate expires, data planes cannot connect to the control plane and stop receiving updates.
23. How do you roll back a bad configuration push?
If you use decK, check out the last known good file from Git, run deck gateway diff to confirm the change set, and run deck gateway sync. Keeping the file as the source of truth makes rollback a normal deployment instead of a manual repair.
24. How do you check a declarative file before loading it into a DB-less node?
Run kong config parse kong.yml, which checks the file's syntax without loading it. In CI, also run deck file validate or deck gateway validate depending on whether you need a running gateway.
25. How do health checks on an upstream work?
Kong supports active checks, where it probes each target on a schedule, and passive checks, where it marks targets unhealthy based on real traffic failures. Unhealthy targets are removed from the load balancer until they recover. You can see the state with GET /upstreams/{upstream}/health.
26. A target is marked unhealthy but the service is fine. How do you bring it back?
Fix the cause first (wrong health check path, timeout too short). With passive checks only, a target may not recover on its own, so mark it healthy through the Admin API endpoint /upstreams/{upstream}/targets/{target}/healthy.
27. What runs Kong on Kubernetes, and how do you check it?
The Kong Ingress Controller translates Ingress, Gateway API and Kong custom resources into Kong configuration. Check its logs for translation errors, and look at Kubernetes events on the resources, since the controller reports invalid objects there.
Kong operations questions for senior engineers
28. What would you alert on for Kong?
A rise in 5xx ratio per service, upstream latency above the SLO, unhealthy upstream targets, database unreachable, data planes not seen by the control plane, config hash drift, and cluster certificate expiry within a set number of days. Page on user impact; send the rest as tickets.
29. How do you plan capacity for Kong?
Measure request rate, latency added by Kong, CPU per node and connection counts under load tests that include the plugins you really run. Plugins such as request transformation, OIDC or rate limiting with a remote counter store change the cost per request, so tests without them mislead.
30. How do you change plugin configuration safely in production?
Apply the change through Git and decK, scope it to one route or consumer first where possible, watch error rate and latency on the dashboard, then widen it. Keep the previous file ready for a fast sync back.
31. How do you protect the Admin API in production?
Keep it on a private interface, put access behind authentication (Kong Enterprise offers RBAC), restrict it with network rules, and prefer declarative changes through CI over manual calls. In hybrid mode, data planes do not serve the Admin API at all.
32. What happens to traffic if the database goes down?
In traditional mode, nodes keep proxying with the configuration cached in memory, but you cannot change configuration and new nodes may fail to start. In hybrid mode, data planes keep their last configuration if the control plane or its database is unavailable. That is one reason hybrid mode is common for production.
33. How do you run a game day for Kong?
In a staging environment, kill a data plane, block the database, expire a test certificate and push a broken route, and check that alerts fire, runbooks work and traffic recovers. Record time to detect and time to recover for each case.
Live incident scenarios
34. Error rate jumps right after a deploy of gateway config. What do you do first?
Roll back with the last known good decK file, then investigate. Compare the diff from the failed pipeline, check the error log for plugin errors, and look at which routes changed status codes.
35. One Kong pod in a Deployment returns 404 for routes that work on other pods. Why?
That pod probably has stale or no configuration, for example a DB-less pod that started before the controller pushed config. Check its readiness and config hash, and confirm that the readiness probe uses /status/ready so such pods get no traffic.
36. Latency rises but kong_upstream_latency_ms is flat. Where do you look?
Inside Kong: plugin cost, DNS resolution, CPU saturation on the nodes, or a remote dependency of a plugin such as an identity provider or Redis for rate limiting. Compare kong_kong_latency_ms by route to find the plugin chain that changed.
37. Clients start getting 429 responses at much lower traffic than expected. What do you check?
The rate limiting plugin's policy and scope. With the local policy each node counts separately, while cluster or redis share counters. A change in node count, a misconfigured limit_by (for example limiting by IP behind a proxy that hides client IPs) or a shared consumer can all cause it.
38. Data planes stop picking up changes. How do you troubleshoot?
Check kong_control_plane_connected on the data planes, the cluster certificate expiry, network reachability to port 8005 on the control plane, and the version compatibility metric. The data plane error log shows connection and TLS errors.
39. You must rotate a TLS certificate on Kong with no downtime. How?
Upload the new certificate and SNI objects through decK or the Admin API before the old one expires. Kong serves the new certificate for new connections without a restart. Verify with openssl s_client against each host name.
40. An upgrade's kong migrations up fails halfway. What now?
Stop the rollout, keep old nodes serving traffic, read the migration error, and check kong migrations list for what was applied. Restore from the pre-upgrade database backup if the documented recovery for that version requires it. This is why the backup comes first in every upgrade plan.
Practise and next steps
Run Kong in Docker with PostgreSQL, enable the Prometheus plugin, scrape it from Prometheus, and practise a drain, an upgrade and a decK rollback. The Kong 4-week study plan gives a schedule, and Kong FAQs covers quick definitions. For broader monitoring, see 10 API monitoring tools.
Gateway operations sit inside wider DevOps work: CI/CD, Kubernetes and monitoring. Our DevOps training covers those skills in the Kharadi, Pune classroom and live online across India.
Frequently asked questions
How do I check if Kong is healthy?
Run kong health on the node or call /status and /status/ready on the Status API.
What does kong drain do?
It makes /status/ready return 503 so load balancers stop sending new traffic before you stop the node.
How do you upgrade Kong with a database?
Back up, run kong migrations up, roll all nodes to the new version, then run kong migrations finish.
What are Kong's default ports?
Proxy 8000 and 8443, Admin API 8001 and 8444 on localhost, and 8005 for the hybrid mode cluster listener.
How do I get Prometheus metrics from Kong?
Enable the Prometheus plugin and scrape /metrics on the Status API, enabling extra metric groups as needed.
How do I roll back Kong configuration?
Sync the last known good declarative file from Git with decK after checking the diff.
Does Kong keep working if PostgreSQL goes down?
Nodes keep proxying with cached configuration, but configuration changes are not possible until the database returns.
Official references
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Angry
0
Sad
0
Wow
0