RDS storage autoscaling not keeping up: how to respond

TL;DR

RDS storage autoscaling grows in fixed steps on a timer, so a sudden write burst can fill the disk faster than the scaler reacts and the instance stalls or crashes. Give the database breathing room immediately by freeing space or scaling up by hand, then set the maximum autoscaling limit and alerts so the next burst does not catch you. Autoscaling is a slow safety net, not burst protection.

RDS storage autoscaling not keeping up

Steps

1. Check current storage use and how fast it is growing.

aws cloudwatch get-metric-statistics --namespace AWS/RDS --metric-name FreeStorageSpace \
  --dimensions Name=DBInstanceIdentifier,Value=[DB_INSTANCE] \
  --start-time [START] --end-time [END] --period 300 --statistics Average,Minimum

Expected: a time series showing free space shrinking. If the slope reaches zero before the next autoscaling step, you need manual action now.

2. Buy time by scaling storage up manually to a size past the immediate burst.

aws rds modify-db-instance --db-instance-identifier [DB_INSTANCE] \
  --allocated-storage [NEW_SIZE_GB] --apply-immediately

Expected: the instance enters a storage-optimization state and free space starts climbing. Writes unblock once free space is comfortably above zero.

3. Free space cheaply: rotate logs, drop old snapshots of the problem, and check for runaway temp tables.

aws rds describe-db-log-files --db-instance-identifier [DB_INSTANCE]

Expected: a list of log files with sizes. Download or purge the large ones, and look for queries writing huge temp tables, which are often the real burst source.

4. Raise the autoscaling ceiling and add a growth-rate alert.

Expected: the autoscaling maximum is set well above the new allocation, and an alarm fires when free space drops below 20 percent or shrinks faster than a threshold per hour. The alert, not the scaler, is what saves you next time.

When to use

• An RDS instance hits storage-full or stalls during a write burst while autoscaling is enabled.
• Free space oscillates near zero even though autoscaling "should" handle it.
• You are reviewing RDS storage settings after any storage-related page.

When not to use

• Storage is steady and the issue is CPU or connections (that is a different bottleneck).
• The instance type does not support autoscaling at all (check the engine and storage type first).
• You are planning initial capacity for a new database (that is sizing, not incident response).

Why it happens

Autoscaling adds storage in increments (10 percent or 10 GB steps depending on configuration) and only evaluates periodically, with a cooldown between actions. A bulk load, a runaway migration, or log spam can consume tens of gigabytes between evaluations. The scaler also stops at the configured maximum, which teams often leave at the default, so growth halts exactly when it is needed most.

Edge cases

• Storage modifications can take a long time on large instances; start the manual scale-up before free space hits zero, not after.
• Some engines require the instance to be in an available state for modifications; if the DB already crashed, restore from snapshot to a larger instance instead.
• Autoscaling never shrinks storage, so a burst permanently raises your bill; right-size down later with a snapshot restore if the growth was one-off.
• io1/io2 volumes and provisioned IOPS interact with storage changes; check IOPS headroom too when writes stall.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_OnY-F0PpEC-fJJUJy0sW8g

Resolved from the public thread: https://vectle.com/posts/pst_OnY-F0PpEC-fJJUJy0sW8g

Thread query: RDS storage autoscaling not keeping up: how to respond

TL;DR: RDS storage autoscaling grows in fixed steps on a timer, so a sudden write burst can fill the disk faster than the scaler reacts and the instance stalls or crashes. Give the database breathing room immediately by freeing space or scaling up by hand, then set the maximum autoscaling limit and alerts so the next burst does not catch you. Autoscaling is a slow safety net, not burst protection.