SSL: error:14094410 – session cache exhaustion in OpenSSL

Direct Answers

1. Automatic lazy expiration? OpenSSL does not implement automatic lazy expiration of stale session-cache entries in any released version. Adding it would require source modifications and rebuilds, introducing eviction-logic complexity and potential performance regressions. The current supported approach is explicit flushing via SSL_CTX_flush_sessions().

2. Cache-size tuning for rapid turnover? For workloads with high connection churn, reduce the cache size (e.g., 256–512 entries via SSL_CTX_sess_set_cache_size()) and schedule frequent flushes (every 5–30 seconds) rather than relying on the default large cache. This limits stale-entry accumulation and makes exhaustion predictable.

3. Pooling and cache-mode interaction? Mixing SSL_SESS_CACHE_CLIENT, SSL_SESS_CACHE_SERVER, and SSL_SESS_CACHE_OFF in the same connection pool causes uneven slot consumption. Keep the cache mode identical across all pooled SSL_CTX objects; otherwise one mode may fill the shared cache while another cannot use it, raising exhaustion risk.

Likely Explanation (Based on Model Knowledge)

The error SSL: error:14094410:SSL routines:ssl3_read_bytes:tlsv1 alert unknown ca appears when the internal session cache is full. OpenSSL's default cache (SSL_SESS_CACHE_SIZE, typically 20480 entries) does not auto-evict expired sessions. When the cache fills, the library cannot store or retrieve a session, falls back to a full handshake, and the peer may reject it with an "unknown CA" alert. This behavior is documented in OpenSSL 3.x internals, but the exact alert code can also stem from certificate verification failures; cache exhaustion is one plausible trigger.

Confirmed Facts (from OpenSSL Documentation)

• SSL_CTX_set_session_cache_mode() controls which side caches sessions.
• SSL_CTX_flush_sessions() must be called by the application to remove expired entries.
• Disabling the cache entirely (SSL_SESS_CACHE_OFF) eliminates resumption benefits and increases handshake latency.

Mitigation Steps

1. Set a smaller cache size early during SSL_CTX initialization:

   SSL_CTX_sess_set_cache_size(ctx, 512);

2. Install a periodic timer (e.g., every 10 s) that calls:

   SSL_CTX_flush_sessions(ctx, time(NULL));

3. Verify all pooled contexts use the same cache mode:

   long mode = SSL_CTX_get_session_cache_mode(ctx);
   assert(mode == DESIRED_MODE);

4. Monitor cache usage via SSL_CTX_sess_number(), SSL_CTX_sess_hits(), and SSL_CTX_sess_misses().

Verification Checklist

• Run a test server with default cache; drive 10,000+ rapid handshakes and confirm the error appears.
• Enable periodic SSL_CTX_flush_sessions() and verify the error stops.
• Reduce cache size to 256 and confirm exhaustion occurs sooner but remains manageable.
• Check SSL_CTX_get_session_cache_mode() and SSL_CTX_get_session_cache_size() across the pool for consistency.

One Missing Diagnostic Detail

What OpenSSL version and exact SSL_CTX_set_session_cache_mode value are you using in the pool? The recommendation changes if you are on OpenSSL 1.1.1 (different default cache size) or if the pool mixes client/server modes.