Suhail Gupta

Software Architect | Mentor
Turning production experience into practical backend & system design insights

[Suhail Gupta @SuhailTechTalks]

@javarevisited Thanks @javarevisited for tagging

[Suhail Gupta @SuhailTechTalks]

I would first check locks- row/table locks or long transactions (pg_locks, pg_stat_activity) can block inserts.
Also things like connection pool limits, WAL/replication lag, or too many indexes can hurt throughput.

We actually faced this in prod- inserts dropped because pgBouncer wasn't utilizing connections properly, even though the app pool was high.
A small config fix there and throughput improved a lot.

[Suhail Gupta @SuhailTechTalks]

@javarevisited CDN only speeds up cached (mostly static) content 🙂

If the website relies on dynamic APIs, backend processing, or heavy JS, it can still be slow.

[Suhail Gupta @SuhailTechTalks]

Firstly I'd ask - do we need avg of the last 1 minute (sliding window) or avg per fixed minute (tumbling window)?
Then keep it simple — maintain a 1-minute window and just track sum + count (small in-memory state).
For each event → update both, and when the window ends → avg = sum / count.
If traffic grows, just partition and aggregate in parallel.

[Suhail Gupta @SuhailTechTalks]

@SumitM_X Sharding by domain may lead to uneven distribution.
Millions of users might share https://t.co/EwqvuiA9Xw.
Hashing the full email would balance shards better.

[Suhail Gupta @SuhailTechTalks]

@vivoplt Since JWTs are stateless, they stay valid until their exp claim expires.
For logout, systems typically expire access tokens quickly and revoke the refresh token.

[Suhail Gupta @SuhailTechTalks]

@javarevisited Usually based on mutual friends and graph connections.
Social networks run recommendation logic on this graph data.

I demonstrated a similar concept in my Redis video.
https://t.co/q1uEG2Ov9n

[Suhail Gupta @SuhailTechTalks]

@coolcoder56 POST triggers a CORS preflight (OPTIONS) request.
If the server doesn't allow or respond to it properly, the browser blocks the request.

[Suhail Gupta @SuhailTechTalks]

@javarevisited Because threads don't run in a fixed order.
The scheduler decides who runs first, so the result can change every time.

I actually demonstrated this in my video on printing 1-N using M threads- shows how important it is to handle this properly.
https://t.co/u5yIrChh4f

[Suhail Gupta @SuhailTechTalks]

K8s is great when different services need different CPU, memory, and scaling policies.

If you put a monolith in K8s, scaling just creates more identical pods of the same monolith, all needing the same resources and carrying the same bottlenecks.

K8s will orchestrate it beautifully- but it’s still one big monolith. 😄

[Suhail Gupta @SuhailTechTalks]

@SumitM_X Both return the same result for an INNER JOIN.
The first uses explicit JOIN syntax, the second uses the old implicit comma join.

The difference?
One clearly defines the relationship… the other just throws the tables together and hopes the WHERE clause behaves. 😄

[Suhail Gupta @SuhailTechTalks]

@SumitM_X Iterator → sequential traversal.
Spliterator → traversal + ability to split the data source for parallel processing.

In short: Iterator walks the collection, Spliterator divides and conquers.

That's why the Java Stream API uses Spliterators internally for parallel streams.

[Suhail Gupta @SuhailTechTalks]

@devops_nk Sharing the in-depth video on How K8s uses etcd, it might help someone-
https://t.co/y9p62QdY1Z

[Suhail Gupta @SuhailTechTalks]

Kafka durability means the message is safe in the log.

Retries exist because everything after poll() — DB writes, API calls, network hops — can still fail.

Kafka protects the data, retries protect the workflow.
Sharing the in-depth Kafka playlist below, it might help someone-
https://t.co/BzGUTXlCxA

[Suhail Gupta @SuhailTechTalks]

Most people using Kubernetes every day have never seen what their cluster state actually looks like.
But every single object in Kubernetes (Pods, Deployments, Secrets, Services…) is stored in etcd.

So what actually happens when you run:
kubectl apply -f pod.yaml
How does that YAML end up inside etcd?

In Part 3 of my ETCD deep dive series, I explain the full flow:
• kubectl → API Server
• API Server → etcd
• How Kubernetes objects are stored
• What the stored data actually looks like
• Explained step-by-step using real Kubernetes YAML

No slides. Just real examples and Kubernetes internals.
If you want to truly understand how Kubernetes works behind the scenes, this video will help.

🎥 Watch here:
https://t.co/TjuqeTya5N
#Kubernetes #DevOps #ETCD #CloudNative #PlatformEngineering

[Suhail Gupta @SuhailTechTalks]

@fromcodetocloud Sharing the in-depth video on how K8s uses etcd, it might help someone-
https://t.co/Nk0VYFggRZ

[Suhail Gupta @SuhailTechTalks]

@asynctrix Sharing the in-depth video on how k8s uses etcd, it might help someone-
https://t.co/Nk0VYFggRZ

[Suhail Gupta @SuhailTechTalks]

@sauravstwt Sharing the in-depth etcd series, it might help someone-
https://t.co/c8sUl8YKIe

[Suhail Gupta @SuhailTechTalks]

@brankopetric00 3 seconds locally. 40 minutes in production.
Looks like schema changes mixed with long-running transactions and concurrent writes caused the trouble here.
There was no rollback plan because AI probably has said: This should work. 😄😀😀

[Suhail Gupta @SuhailTechTalks]

@brankopetric00 In other words, you have one environment ready to disappoint users.😂😂😂