## TL;DR
AWS processes one modification per reservation at a time and caps how many new reservations a single request can create. Check for a pending modification first and wait it out; otherwise split into fewer pieces per request and retry. The original reservation keeps its end date through every split.

```text
ReservedInstancesModificationLimitExceeded
```

## Steps
1. Check for a pending modification: `aws ec2 describe-reserved-instances-modifications --filters Name=status,Values=pending`.
   Expected: if a modification is pending, that is the blocker - you cannot stack another one on the same reservation.
2. If one is pending, wait for it to reach completed or failed status, then resubmit. Do not cancel it; a pending request cannot be changed or canceled after submission.
   Expected: status moves off pending and the reservation accepts a new request.
3. If nothing is pending, reduce the number of target configurations in the split - fewer new reservations per request.
   Expected: the smaller request is accepted.
4. For automation, make splits sequential: submit one request, poll the modification status to completed, then submit the next.
   Expected: no more limit errors across the batch.
5. Verify the result: the original reservation shows retired and the new reservations are active with the same end date.
   Expected: capacity is redistributed and discounts apply from the hour of modification.

## Use this when
- Splitting one RI into several smaller ones fails.
- An automation loop that rebalances RIs across AZs starts erroring.
- A modification was just submitted and the follow-up request is rejected.

## Not for this skill when
- The error is the requested-configuration-not-supported message - that is a wrong-attribute problem.
- Convertible RI exchanges fail - exchanges have their own value rules.

## Variant phrasings
- Cannot split reserved instance modification limit
- EC2 modify reserved instances limit exceeded
- Split RI into multiple reservations fails

## Why it happens
Each modification rewrites the billing linkage for the reservation, so AWS serializes modifications per reservation and bounds the fan-out of a single request to keep the accounting deterministic. Rapid-fire splits from automation hit the pending-request gate most often.

## Edge cases
- Every resulting reservation keeps the original end date; splitting does not extend the term.
- Tiny splits can strand capacity in AZs with no matching usage - size the splits to the real AZ distribution.
- Modification is free and failed requests leave the original untouched, so a rejected split is safe to retry in smaller pieces.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_a9BBlE6tWo6zX82dd_ae6w
