9.2 KiB

Raw Blame History

Distributed IAM for SeaweedFS S3 Gateway

This document explains how to configure SeaweedFS S3 Gateway for distributed environments with multiple gateway instances.

Problem Statement

In distributed environments with multiple S3 gateway instances, the default in-memory storage for IAM components causes serious consistency issues:

Roles: Each instance maintains separate in-memory role definitions
Sessions: STS sessions created on one instance aren't visible on others
Policies: Policy updates don't propagate across instances
Authentication failures: Users may get different responses from different instances

Solution: Filer-Based Distributed Storage

SeaweedFS now supports distributed IAM storage using the filer as a centralized backend for:

✅ Role definitions (FilerRoleStore)
✅ STS sessions (FilerSessionStore)
✅ IAM policies (FilerPolicyStore)

All S3 gateway instances share the same IAM state through the filer.

Configuration

Standard Configuration (Single Instance)

{
  "sts": {
    "tokenDuration": 3600000000000,
    "maxSessionLength": 43200000000000,
    "issuer": "seaweedfs-sts",
    "signingKey": "base64-encoded-signing-key"
  },
  "policy": {
    "defaultEffect": "Deny",
    "storeType": "memory"
  }
}

Distributed Configuration (Multiple Instances)

{
  "sts": {
    "tokenDuration": 3600000000000,
    "maxSessionLength": 43200000000000,
    "issuer": "seaweedfs-sts", 
    "signingKey": "base64-encoded-signing-key",
    "sessionStoreType": "filer",
    "sessionStoreConfig": {
      "filerAddress": "localhost:8888",
      "basePath": "/seaweedfs/iam/sessions"
    }
  },
  "policy": {
    "defaultEffect": "Deny",
    "storeType": "filer",
    "storeConfig": {
      "filerAddress": "localhost:8888",
      "basePath": "/seaweedfs/iam/policies"
    }
  },
  "roleStore": {
    "storeType": "filer",
    "storeConfig": {
      "filerAddress": "localhost:8888", 
      "basePath": "/seaweedfs/iam/roles"
    }
  }
}

Storage Backends

Memory Storage (Default)

Performance: Fastest access (sub-millisecond)
Persistence: Lost on restart
Distribution: ❌ Not shared across instances
Use case: Single instance or development

Filer Storage (Distributed)

Performance: Network latency + filer performance
Persistence: ✅ Survives restarts
Distribution: ✅ Shared across all instances
Use case: Production multi-instance deployments

Deployment Examples

Single Instance (Simple)

# Standard deployment - no special configuration needed
weed s3 -filer=localhost:8888 -port=8333

Multiple Instances (Distributed)

# Instance 1
weed s3 -filer=localhost:8888 -port=8333 -iam.config=/path/to/distributed_config.json

# Instance 2  
weed s3 -filer=localhost:8888 -port=8334 -iam.config=/path/to/distributed_config.json

# Instance N
weed s3 -filer=localhost:8888 -port=833N -iam.config=/path/to/distributed_config.json

All instances share the same IAM state through the filer.

Load Balancer Configuration

upstream seaweedfs-s3 {
    # No need for session affinity with distributed storage
    server gateway-1:8333;
    server gateway-2:8334; 
    server gateway-3:8335;
}

server {
    listen 80;
    location / {
        proxy_pass http://seaweedfs-s3;
    }
}

Storage Locations

When using filer storage, IAM data is stored at:

/seaweedfs/iam/
├── sessions/           # STS session tokens
│   ├── session_abc123.json
│   └── session_def456.json
├── policies/           # IAM policy documents  
│   ├── policy_S3AdminPolicy.json
│   └── policy_S3ReadOnlyPolicy.json
└── roles/              # IAM role definitions
    ├── S3AdminRole.json
    └── S3ReadOnlyRole.json

Performance Considerations

Memory vs Filer Storage

Aspect	Memory	Filer
Read Latency	<1ms	~5-20ms
Write Latency	<1ms	~10-50ms
Consistency	Per-instance	Global
Persistence	None	Full
Scalability	Limited	Unlimited

Optimization Tips

Use fast storage for the filer (SSD recommended)
Co-locate filer with S3 gateways to reduce network latency
Enable filer caching for frequently accessed IAM data
Monitor filer performance - IAM operations depend on it

Migration Guide

From Single to Multi-Instance

Stop existing S3 gateway

Update configuration to use filer storage:

{
  "policy": { "storeType": "filer" },
  "sts": { "sessionStoreType": "filer" },
  "roleStore": { "storeType": "filer" }
}

Start first instance with new config
Verify IAM data was migrated to filer
Start additional instances with same config

From Memory to Filer Storage

IAM data in memory is not automatically migrated. You'll need to:

Export existing roles/policies from configuration files
Update configuration to use filer storage
Restart with new config - data will be loaded from config into filer
Remove config-based definitions (now stored in filer)

Troubleshooting

Configuration Issues

# Check IAM configuration is loaded correctly
grep "advanced IAM" /path/to/s3-gateway.log

# Verify filer connectivity
weed filer.cat /seaweedfs/iam/roles/

Inconsistent Behavior

# Check if all instances use same filer
curl http://gateway-1:8333/status  
curl http://gateway-2:8334/status

# Verify IAM storage locations exist
weed filer.ls /seaweedfs/iam/

Performance Issues

# Monitor filer latency
curl http://filer:8888/stats

# Check IAM storage path performance  
time weed filer.cat /seaweedfs/iam/roles/TestRole.json

Best Practices

Security

Secure filer access - use TLS and authentication
Limit IAM path access - only S3 gateways should access /seaweedfs/iam/
Regular backups of IAM data in filer
Monitor access to IAM storage paths

High Availability

Replicate filer across multiple nodes
Use fast, reliable storage for filer backend
Monitor filer health - critical for IAM operations
Implement alerts for IAM storage issues

Capacity Planning

Estimate IAM data size: Roles + Policies + Sessions
Plan for growth: Active sessions scale with user count
Monitor filer disk usage: /seaweedfs/iam/ path
Set up log rotation: For IAM audit logs

Architecture Comparison

Before (Memory-Only)

┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│ S3 Gateway 1│    │ S3 Gateway 2│    │ S3 Gateway 3│
│             │    │             │    │             │  
│ ┌─────────┐ │    │ ┌─────────┐ │    │ ┌─────────┐ │
│ │ IAM     │ │    │ │ IAM     │ │    │ │ IAM     │ │
│ │ Memory  │ │    │ │ Memory  │ │    │ │ Memory  │ │
│ └─────────┘ │    │ └─────────┘ │    │ └─────────┘ │
└─────────────┘    └─────────────┘    └─────────────┘
      ❌               ❌               ❌
 Inconsistent      Inconsistent    Inconsistent

After (Filer-Distributed)

┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│ S3 Gateway 1│    │ S3 Gateway 2│    │ S3 Gateway 3│
│             │    │             │    │             │
└──────┬──────┘    └──────┬──────┘    └──────┬──────┘
       │                  │                  │
       └──────────────────┼──────────────────┘
                          │
                ┌─────────▼─────────┐
                │   SeaweedFS       │
                │   Filer           │
                │ ┌───────────────┐ │
                │ │ /seaweedfs/   │ │
                │ │ iam/          │ │
                │ │ ├─roles/      │ │
                │ │ ├─policies/   │ │
                │ │ └─sessions/   │ │
                │ └───────────────┘ │
                └───────────────────┘
                        ✅
                   Consistent

Summary

The distributed IAM system solves critical consistency issues in multi-instance deployments by:

Centralizing IAM state in SeaweedFS filer
Ensuring consistency across all gateway instances
Maintaining performance with configurable storage backends
Supporting scalability for any number of gateway instances

For production deployments with multiple S3 gateway instances, always use filer-based storage for all IAM components.

9.2 KiB Raw Blame History