Amazon S3 Glacier Storage Classes Explained

Simple explanation

Glacier is not a separate browsable filesystem. It is long-term, low-cost storage inside the same S3 service you already use, accessed through the same S3 API and console.

The distinction that trips most people up is which classes require a restore request and which do not:

• S3 Glacier Instant Retrieval: data is available in milliseconds, no restore step needed
• S3 Glacier Flexible Retrieval: data is archived; you submit a restore request and wait minutes to hours
• S3 Glacier Deep Archive: data is archived; restore takes 12 to 48 hours

How Amazon S3 Glacier works

Every S3 object belongs to exactly one storage class. You can set the Glacier class at upload, or let S3 lifecycle policies transition objects into colder classes automatically as they age.

Glacier Instant Retrieval stores data in a way that allows millisecond access, similar to Standard-IA. No restore request is needed. The retrieval fee per GB is higher than Standard-IA, which is the trade-off for the lower storage cost.

Glacier Flexible Retrieval archives objects so they are not directly readable. To access one, you submit a restore request and choose a retrieval tier (Expedited, Standard, or Bulk). The restore process creates a temporary S3 Standard copy for a number of days you specify, after which the copy expires and the object returns to archived status.

Glacier Deep Archive works the same way as Flexible Retrieval but with the lowest storage cost and the slowest restore. Only Standard (12 hours) and Bulk (48 hours) restore tiers are available. There is no Expedited option.

Minimum storage durations affect your bill if you delete or transition objects too early. Glacier Instant Retrieval and Flexible Retrieval both have 90-day minimums. Deep Archive has a 180-day minimum. Delete an object on day 30 and AWS still charges you for the full minimum period at that class’s rate.

Lifecycle transitions are the practical approach for moving data into Glacier at scale. Rather than uploading directly to a Glacier class, you write a rule once and S3 applies it automatically. See S3 Lifecycle Policies for configuration details, including minimum age gaps between transitions.

Object Lock for compliance archives. S3 Object Lock in Compliance mode prevents any user, including the AWS root account, from deleting or modifying an object before its retention period expires. Glacier Deep Archive combined with Object Lock is a common pattern for seven-year regulatory archives under rules like SEC Rule 17a-4, HIPAA, and FINRA.

S3 Glacier storage classes comparison

Which Glacier class should you choose?

Choose Glacier Instant Retrieval if…#

• The data is archived but might need to be accessed without advance notice
• A delay of even a few minutes would disrupt a workflow or user-facing process
• You access the data less than once per quarter but cannot predict exactly when
• Examples: medical imaging archives (a clinician requests a five-year-old scan mid-appointment), legal case files that must be produced quickly on demand, customer contract archives retrieved during disputes

Choose Glacier Flexible Retrieval if…#

• The data is accessed at most once or twice a year
• You can plan the retrieval in advance and a wait of minutes to hours is acceptable
• You want lower storage costs than Instant Retrieval and are willing to trade away immediate availability
• Examples: annual audit records, year-end financial data, completed project deliverables, historical log archives reviewed on a set schedule

Choose Glacier Deep Archive if…#

• The data must be retained for legal or regulatory reasons, typically 7 or more years
• You almost never expect to retrieve it; access is genuinely the exception, not the rule
• A 12 to 48 hour retrieval window is completely acceptable for your scenario
• You have confirmed that retrieval costs at your expected scale are budgeted
• Examples: long-term compliance records (SEC, HIPAA, GDPR), archived transaction histories, old database dumps held under legal hold, decommissioned system backups kept for contractual obligations

Glacier vs Standard-IA vs Intelligent-Tiering

The right choice between Glacier and the cheaper non-Glacier classes comes down to two questions: how fast do you need access, and how predictable is the access pattern?

When Standard-IA is better than Glacier: if data is accessed a handful of times per year but the timing is unpredictable, Standard-IA gives you instant access with a reasonable retrieval fee, no 90-day minimum, and simpler lifecycle management.

When Glacier Instant Retrieval beats Standard-IA: for objects stored well beyond 90 days and accessed very infrequently, Instant Retrieval’s lower per-GB storage cost outweighs Standard-IA’s lower retrieval fee. If you access data less than once per quarter and hold it for a year or more, Instant Retrieval typically wins on total cost.

When Intelligent-Tiering is the better fit: if access patterns are genuinely unpredictable, with some objects staying hot and others going cold immediately, Intelligent-Tiering handles it automatically. It has no retrieval fees and no minimum storage duration charges, but it does charge a small per-object monitoring fee that makes it poor value for tiny or very short-lived objects.

When Deep Archive is the wrong choice entirely: if there is any realistic chance the data could be needed on short notice, or if a large unexpected retrieval would generate a bill you have not planned for, Deep Archive is not the right call.

How to move data into Glacier

There are two ways to get objects into a Glacier storage class.

Option 1: Direct upload. When using the AWS CLI or an SDK, specify the storage class at upload time:

# Upload directly to Glacier Instant Retrieval
aws s3 cp report.pdf s3://my-archive-bucket/reports/report.pdf \
  --storage-class GLACIER_IR

# Upload directly to Glacier Deep Archive
aws s3 cp annual-records.zip s3://my-archive-bucket/records/annual-records.zip \
  --storage-class DEEP_ARCHIVE

Option 2: Lifecycle transitions. For large volumes, lifecycle policies are the practical approach. Write a rule once and S3 transitions objects automatically as they age:

{
  "Rules": [
    {
      "ID": "archive-old-reports",
      "Status": "Enabled",
      "Filter": { "Prefix": "reports/" },
      "Transitions": [
        { "Days": 90, "StorageClass": "GLACIER_IR" },
        { "Days": 365, "StorageClass": "DEEP_ARCHIVE" }
      ]
    }
  ]
}

This rule moves objects in the reports/ prefix to Glacier Instant Retrieval after 90 days and to Deep Archive after 365 days. See S3 Lifecycle Policies for the full guide, including how transitions interact with S3 versioning and the minimum age gaps AWS requires between transitions.

How to restore archived data

How you access Glacier data depends entirely on which class it is in.

Glacier Instant Retrieval: objects are accessible immediately via a standard S3 GET request. No restore step is required.

Glacier Flexible Retrieval and Deep Archive: these classes archive the object in a way that makes it unreadable until you submit a restore request. The restore process copies the object to a temporary S3 Standard location for a configurable number of days, after which the copy expires and the object returns to archived status.

Glacier Flexible Retrieval offers three restore tiers:

Deep Archive only supports Standard (12 hours) and Bulk (48 hours). There is no Expedited option for Deep Archive.

Initiating a restore with the AWS CLI:

# Restore a Glacier Flexible Retrieval object using Standard tier
# Keep the restored copy available for 7 days
aws s3api restore-object \
  --bucket my-archive-bucket \
  --key records/2024/annual-report.pdf \
  --restore-request '{"Days": 7, "GlacierJobParameters": {"Tier": "Standard"}}'

After the restore completes, you download the object via a normal S3 GET. After 7 days the temporary copy expires. The original archived object remains in place throughout.

Provisioned retrieval capacity is an option for Glacier Flexible Retrieval if Expedited restores are business-critical. Purchasing provisioned capacity guarantees Expedited requests succeed even during peak demand. Without it, Expedited restores can be rate-limited. If your workflow requires reliably fast restores, provisioned capacity is worth the fixed hourly cost.

When not to use Glacier

Glacier is the right tool for cold data held long-term. It is the wrong tool in several common situations.

Disaster recovery with a short RTO. If your recovery plan requires restoring from backup within an hour, Flexible Retrieval and Deep Archive cannot meet that objective. Keep DR backups in Standard-IA or Glacier Instant Retrieval, and confirm the class you choose aligns with your actual recovery time objective. See Choosing the Right Storage Service for broader storage design decisions.

Frequently accessed application data. If data is being read monthly, weekly, or more often, retrieval fees in any Glacier class will erode the storage savings. With Flexible Retrieval or Deep Archive, the restore delays will break application workflows entirely. Standard or Standard-IA is a far better fit.

Unpredictable retrieval-heavy workloads. If the access pattern is unknown and data could be needed at any time by any number of users, the risk of unexpected restore delays and retrieval fees is too high. Intelligent-Tiering handles unknown access patterns more gracefully and has no retrieval fees.

Short-lived data. If an object will be deleted within weeks or a few months, the minimum storage duration charges in Glacier classes (90 or 180 days) can make Glacier more expensive than Standard. Never move short-lived data to Glacier.

Workloads with access SLAs. If users or systems expect data to be available in under a second, Flexible Retrieval and Deep Archive are not viable options. Even Instant Retrieval carries a higher retrieval fee than Standard-IA if access frequency is higher than expected.

Object Lock for compliance archives

S3 Object Lock in Compliance mode provides WORM (Write Once, Read Many) protection for archived objects. Once applied, the lock cannot be removed by anyone, including the AWS root account, until the retention period expires.

This is the standard mechanism for regulatory archives subject to rules like SEC Rule 17a-4, HIPAA, and FINRA. Glacier Deep Archive combined with Object Lock creates low-cost archives that cannot be tampered with or deleted before the retention date.

# Store an object in Glacier Deep Archive with a 7-year compliance lock
aws s3api put-object \
  --bucket my-compliance-archive \
  --key regulatory/2026/annual-report.pdf \
  --body annual-report.pdf \
  --storage-class DEEP_ARCHIVE \
  --object-lock-mode COMPLIANCE \
  --object-lock-retain-until-date 2033-01-01T00:00:00Z

Once this lock is applied, the object cannot be deleted until 2033 by anyone. For guidance on S3 access controls and Object Lock configuration, including bucket policies and encryption for sensitive archives, see S3 Security Best Practices.