Mudda

Building CostPatrol — AWS cost optimisation that finds waste in minutes. Scan findings, AWS gotchas, FinOps takes. Software engineer.

[Mudda @muddawy]

Scanned an AWS account this morning.
CostPatrol found $1,238 - $1,537/mo in waste.

4 minutes. 56 findings. 17 rule types. 7 regions.

Biggest categories: redundant NAT gateways, idle EC2, RDS on previous-gen hardware.

If your AWS bill feels wrong, it probably is. https://t.co/lrO9zn5ylg

[Mudda @muddawy]

@visafreeworld You forgot about Dahab, Egypt

[Mudda @muddawy]

S3 Bucket Keys only cover objects uploaded after you switch it on.

Everything already in the bucket keeps paying a KMS call on every read until you copy each object over itself, which resets its age for lifecycle rules.

Free scan at https://t.co/eljSIDKVHn

[Mudda @muddawy]

Worth knowing before you migrate: hold Redis OSS reservations, move those nodes to Valkey, and the reservations carry over. Same node type family, an r7g stays an r7g, same region.

Moving to Valkey doesn't cost you the discount you already bought.

Free scan at https://t.co/qtFwl1aMFn

[Mudda @muddawy]

Database Savings Plans landed in December 2025. Up to 35% off if you commit to a fixed hourly spend for a year.

The headline doesn't mention the floor: coverage is Generation 7 and newer instances only. A fleet on r5 or r6g gets nothing.

[Mudda @muddawy]

The ElastiCache entry is narrower than it looks. AWS limits it to Valkey.

A fleet on Redis OSS or Memcached isn't on that list. Redis OSS buys reserved nodes instead, on its own one or three year clock.

[Mudda @muddawy]

I've put off a version upgrade too. This one runs a meter while you decide.

Two ways off it on a running database: upgrade the engine or delete it.

Free scan at https://t.co/eljSIDKVHn

[Mudda @muddawy]

MySQL 5.7 on RDS didn't stop working when standard support ended in Feb 2024. It carried on, and AWS has been billing Extended Support on top ever since: an hourly charge for every CPU the instance has.

On 1 March 2026 that rate doubled. Nobody touched the database.

[Mudda @muddawy]

Reserved Instances don't cover it. Your instance price can be locked in for three years and this surcharge still lands in full, outside the reservation.

Multi-AZ and read replicas count too. The standby has CPUs, so that db.r5.xlarge is billed for 8 vCPUs, not 4.

[Mudda @muddawy]

The fix is a redeploy per service. With three zones and round-robin, about two-thirds of those Service Connect calls leave the zone today, and zone-aware routing pulls most of them back.

Free scan at https://t.co/eljSIDKVHn

[Mudda @muddawy]

Two ECS tasks talking in the same zone, private IPs: free. The same call to a task in a neighbouring zone: $0.01/GB leaving the sender, $0.01/GB arriving at the receiver.

25 TB a month across that boundary costs about $500. Straight off the price list, not a scan.

[Mudda @muddawy]

Nothing tells you if a service is still on the old behaviour. There's no alert for it.

You get one regional data transfer line under EC2. It tells you the region, not which service made the calls.

[Mudda @muddawy]

@lightsilver323 https://t.co/eljSIDKVHn

[Mudda @muddawy]

@sneako I love you brother

[Mudda @muddawy]

Moved our ECS Fargate to Graviton, ~20% off compute. Already multi-arch? Just a flag flip. We weren't, and a native dep with no arm64 build was the whole migration.

Free scan at https://t.co/eljSIDKVHn

[JonnyUtd @Fx1Jonny]

This guy was killed defending his family and his home.

He did the same thing any 1 of us men would do for our families. RIP Sir x

That’s heroism.