Deploying in a Production Environment

For production environments, we recommend that you optimize PostgreSQL settings and configure automated maintenance. This guide covers configuration, monitoring, and high availability considerations for production deployments.

PostgreSQL Configuration

You should optimize PostgreSQL memory and performance settings for semantic caching workloads. Proper configuration ensures optimal cache performance and efficient resource utilization.

In the following example, the ALTER SYSTEM commands configure PostgreSQL memory settings for semantic caching workloads:

ALTER SYSTEM SET shared_buffers = '4GB';
ALTER SYSTEM SET effective_cache_size = '12GB';
ALTER SYSTEM SET work_mem = '256MB';

SELECT pg_reload_conf();

Adjust the shared_buffers setting based on your available RAM. The effective_cache_size should typically be 50-75% of total RAM. The work_mem setting allocates memory for vector operations.

Automated Maintenance

You can schedule automatic cache maintenance tasks using the pg_cron extension. Regular maintenance prevents cache bloat and ensures optimal performance.

In the following example, the cron.schedule() function sets up automatic cache maintenance tasks:

CREATE EXTENSION IF NOT EXISTS pg_cron;

SELECT cron.schedule(
    'semantic-cache-eviction',
    '*/15 * * * *',
    $$SELECT semantic_cache.auto_evict()$$
);

SELECT cron.schedule(
    'semantic-cache-cleanup',
    '0 * * * *',
    $$SELECT semantic_cache.evict_expired()$$
);

SELECT * FROM cron.job WHERE jobname LIKE 'semantic-cache%';

The first job runs auto-eviction every 15 minutes. The second job removes expired entries every hour.

Index Optimization

Choose the appropriate vector index strategy based on your cache size. Different index types provide optimal performance at different scales.

Small to Medium Caches

The default IVFFlat index works well for caches with fewer than 100,000 entries. No additional configuration is required for this cache size.

Large Caches

For caches containing between 100,000 and 1 million entries, increase the IVFFlat lists parameter for better performance.

In the following example, the CREATE INDEX command creates an optimized IVFFlat index for large caches:

DROP INDEX IF EXISTS semantic_cache.idx_cache_embedding;
CREATE INDEX idx_cache_embedding
    ON semantic_cache.cache_entries
    USING ivfflat (query_embedding vector_cosine_ops)
    WITH (lists = 1000);

Very Large Caches

For caches exceeding 1 million entries, use the HNSW index for optimal performance. The HNSW index requires pgvector version 0.5.0 or later.

In the following example, the CREATE INDEX command creates an HNSW index for very large caches:

DROP INDEX IF EXISTS semantic_cache.idx_cache_embedding;
CREATE INDEX idx_cache_embedding_hnsw
    ON semantic_cache.cache_entries
    USING hnsw (query_embedding vector_cosine_ops)
    WITH (m = 16, ef_construction = 64);

The HNSW index provides the following benefits:

• The HNSW index delivers faster queries with 1-2ms response times.
• HNSW provides better recall accuracy at high similarity thresholds.
• HNSW scales linearly with cache size for consistent performance.

Configuring Monitoring

You can configure custom views to monitor cache health and performance metrics. Regular monitoring helps identify performance issues and optimize cache configuration.

In the following example, the CREATE VIEW command creates a production monitoring dashboard:

CREATE OR REPLACE VIEW semantic_cache.production_dashboard AS
SELECT
    (SELECT hit_rate_percent FROM semantic_cache.cache_stats())::NUMERIC(5,2) || '%' AS hit_rate,
    (SELECT total_entries FROM semantic_cache.cache_stats()) AS total_entries,
    (SELECT pg_size_pretty(SUM(result_size_bytes)::BIGINT) FROM semantic_cache.cache_entries) AS cache_size,
    (SELECT COUNT(*) FROM semantic_cache.cache_entries WHERE expires_at <= NOW()) AS expired_entries,
    (SELECT value FROM semantic_cache.cache_config WHERE key = 'eviction_policy') AS eviction_policy,
    NOW() AS snapshot_time;

SELECT * FROM semantic_cache.production_dashboard;

High Availability Considerations

The cache integrates seamlessly with PostgreSQL's replication and backup mechanisms. The semantic cache data automatically replicates with standard PostgreSQL streaming replication.

In the following example, the pg_dump command creates a backup of cache metadata:

pg_dump -U postgres -d your_db \
    -t semantic_cache.cache_entries \
    -t semantic_cache.cache_metadata \
    -F c -f cache_backup.dump

The cache data automatically replicates with PostgreSQL streaming replication. No special configuration is needed for replication.