What's the difference between Standard-IA and One Zone-IA?

Standard-IA replicates across at least 3 Availability Zones, same as Standard. One Zone-IA stores data in only a single AZ at a lower price, but the data is lost if that AZ is destroyed — appropriate only for easily-reproducible data like secondary backups.

Can lifecycle rules apply to specific tags instead of prefixes?

Yes — a lifecycle rule's Filter can match by object tag, prefix, or both combined, which is useful when objects needing different retention live in the same prefix and can't be distinguished by path alone.

S3 Storage Classes & Lifecycle Policies

Section 01

Prerequisites

An AWS account
An existing S3 bucket, or the ability to create one

Section 02

What You'll Learn

Compare S3 Standard, Standard-IA, Intelligent-Tiering, Glacier, and Deep Archive on cost and retrieval latency
Write a lifecycle rule that transitions objects between storage classes by age
Configure a rule that expires objects and incomplete multipart uploads automatically
Choose the correct storage class for a given access pattern

Section 03

Theory

S3 charges for three separate things: storage (per GB/month), requests (per GET/PUT), and retrieval (for the colder tiers only). Storage classes trade a lower per-GB storage price for a higher retrieval cost and/or longer retrieval latency — the right choice depends entirely on how often, and how urgently, the data is actually read back.

Storage class	Retrieval time	Best for
Standard	Milliseconds	Actively accessed data
Intelligent-Tiering	Milliseconds	Unpredictable/unknown access patterns
Standard-IA / One Zone-IA	Milliseconds	Infrequent access, but instant when needed
Glacier Instant Retrieval	Milliseconds	Archive data rarely read but needed instantly
Glacier Flexible Retrieval	Minutes to hours	Archive, retrieval isn't time-critical
Glacier Deep Archive	Up to ~12 hours	Long-term compliance archives, rarely if ever read

Intelligent-Tiering is worth calling out specifically: it monitors each object's access pattern and automatically moves it between frequent and infrequent tiers with no retrieval fee and no performance impact — the right default when you genuinely don't know the access pattern in advance.

Section 04

Architecture

A typical compliance-log lifecycle: hot for 30 days, cheaper-but-instant for 60 more, deep archive for a year, then deleted

Section 05

Hands-On Lab

lifecycle-policy.json

{
  "Rules": [
    {
      "ID": "log-tiering",
      "Filter": { "Prefix": "logs/" },
      "Status": "Enabled",
      "Transitions": [
        { "Days": 30, "StorageClass": "STANDARD_IA" },
        { "Days": 90, "StorageClass": "GLACIER" },
        { "Days": 365, "StorageClass": "DEEP_ARCHIVE" }
      ],
      "Expiration": { "Days": 1825 },
      "AbortIncompleteMultipartUpload": { "DaysAfterInitiation": 7 }
    }
  ]
}

aws s3api put-bucket-lifecycle-configuration \
  --bucket my-app-logs \
  --lifecycle-configuration file://lifecycle-policy.json

# Verify it applied
aws s3api get-bucket-lifecycle-configuration --bucket my-app-logs

# Check which storage class an object is currently in
aws s3api head-object --bucket my-app-logs --key logs/2026-01-01.log \
  --query StorageClass

Section 07

Best Practices

Always set AbortIncompleteMultipartUpload

Failed or abandoned multipart uploads leave orphaned parts that silently accrue storage charges forever unless this rule cleans them up — it costs nothing to enable and has no downside.

Use Storage Class Analysis before writing transition rules

S3's built-in Storage Class Analysis observes actual access patterns for 30+ days and recommends transition ages — use it instead of guessing at day thresholds for an existing bucket.

Section 08

Common Mistakes

Transitioning small objects to Glacier

Objects under 128KB aren't worth transitioning — Glacier tiers charge a per-object overhead that can exceed the storage savings for small files. Filter transitions to larger objects, or exclude small-object prefixes.

Forgetting IA's per-GB retrieval fee when planning access

Standard-IA and One Zone-IA are cheaper to store but charge a retrieval fee per GB read — a dataset read back frequently despite being tiered to IA can end up costing more than leaving it in Standard. Match the class to actual read frequency, not just age.

Section 09

Troubleshooting

Lifecycle rule isn't transitioning objects:rules only evaluate objects going forward from when the rule was created for age-based transitions in some edge cases with versioning — confirm the rule's Filter (prefix/tag) actually matches the target objects, and note that transitions run as a daily batch process, not instantly at the exact day boundary.

Trying to read a Glacier object returns an error, not data: objects in Glacier Flexible Retrieval or Deep Archive must be explicitly restored first (aws s3api restore-object) before a GET succeeds — a direct GetObject against an un-restored archived object fails with an InvalidObjectState error.

Section 10

Interview Questions

S3 Standard for frequently accessed data; Standard-IA/One Zone-IA for infrequent access with a retrieval fee (cheaper storage, costs more per GB retrieved); Intelligent-Tiering when access patterns are unpredictable (it auto-moves objects between tiers); Glacier Instant/Flexible/Deep Archive for long-term archival where retrieval can take minutes to hours in exchange for the lowest storage cost. The decision is driven by how often the data is actually read and how quickly you'd need it back if you did.

See all 20 AWS interview questions

FAQ

Frequently Asked Questions

No — the object key and URL stay identical. The storage class only affects cost and (for Glacier tiers) whether a restore step is needed before the object can be read.

Section 12

Summary

Choose a storage class based on actual read frequency and urgency, not just object age; Intelligent-Tiering is the safe default for unknown access patterns; lifecycle rules automate transitions and expiration so nobody has to manually tier or delete objects; and Glacier-tier objects require an explicit restore before they can be read.