Why Fast APIs Crash with 'Cannot Assign Address' (TIME_WAIT)

🎯 The Question

"You deploy a high-performance backend microservice making 5,000 HTTP requests per second to an external payment API. Under load, your service suddenly crashes with the error: EADDRNOTAVAIL: Cannot assign requested address. Why does closing TCP connections trigger this crash, and what is the TIME_WAIT state?"

---

⚡ 30-Second Elevator Pitch

Every outgoing TCP connection is uniquely identified by a 4-Tuple: $\{\text{Source IP}, \ \mathbf{Source\ Port}, \ \text{Destination IP}, \ \text{Destination\ Port}\}$

When your service connects to an external API (api.stripe.com:443), three of those variables are fixed. The Linux kernel must assign an unused Ephemeral Port (by default, ports 32768 to 60999, offering $\approx \mathbf{28{,}232\text{ ports}}$).

The TIME_WAIT Trap:

• Whichever side actively closes the TCP connection first must stay in the TIME_WAIT state for $2 \times \text{MSL}$ (Maximum Segment Lifetime = 60 seconds).
• During these 60 seconds, that local port number is permanently locked by the kernel to ensure lingering packets from the old connection do not corrupt future connections.
• At 5,000 req/sec without connection pooling: $5{,}000 \times 60\text{ seconds} = \mathbf{300{,}000\text{ ports needed}}$
• The system exhausts all 28,000 ephemeral ports in under 6 seconds, throwing Cannot assign requested address!

---

🧠 Under-the-Hood: Ephemeral Port Exhaustion Pipeline

---

🔬 How to Fix: HTTP Keep-Alive & Connection Pooling

The primary production solution is to stop creating and destroying TCP connections:

1. Enable HTTP Keep-Alive & Connection Pooling

Reuse existing TCP connections across thousands of requests:

• Node.js: Use http.Agent({ keepAlive: true, maxSockets: 500 }).
• Go: http.Client with Transport.MaxIdleConnsPerHost = 100.
• Python: Use requests.Session() instead of raw requests.get().

2. Linux Kernel Socket Reuse (tcp_tw_reuse)

Allow the kernel to safely recycle TIME_WAIT sockets for outgoing connections when timestamps match:

sysctl -w net.ipv4.tcp_tw_reuse=1

3. Expand Ephemeral Port Range

sysctl -w net.ipv4.ip_local_port_range="1024 65535"

---

📌 Comparison Matrix: Unpooled Requests vs. Connection Pooling

Metric	Without Connection Pool (Short-Lived Sockets)	With HTTP Connection Pool (Persistent Keep-Alive)
TCP Handshake Overhead	3-way handshake on every single request	Handshake done once; reused for thousands of requests
TLS / SSL Handshake	Expensive crypto negotiation on every call	0 ms overhead after initial connection
Ephemeral Port Churn	Thousands of ports consumed per second	Uses a fixed pool of ~50 to 500 socket ports
TIME_WAIT Accumulation	Rapidly exhausts port limits (EADDRNOTAVAIL)	Near zero sockets in TIME_WAIT state
API Latency	50–150 ms per request	2–5 ms per request ($20\times$ faster!)

---

💡 What Interviewers Ask Next (Follow-Up Traps)

1. "Why does TCP mandate the TIME_WAIT state in the first place?"

   • Answer: Two critical reasons:
     1. Reliable connection termination: If the final ACK from the active closer is lost, the peer retransmits FIN. The closer must be in TIME_WAIT to resend ACK, rather than responding with RST.
     2. Preventing packet corruption: Lingering duplicate packets delayed in routers must completely die in the network before a new connection reuses the same 4-tuple.

2. "Why was net.ipv4.tcp_tw_recycle removed from recent Linux kernels?"

   • Answer: tcp_tw_recycle aggressively recycled sockets based on per-host timestamps. When clients sat behind a NAT gateway (like mobile carriers or corporate firewalls), multiple devices shared one public IP with unsynchronized clock timestamps, causing legitimate connections to be dropped randomly. It was removed in Linux 4.12.

---

Placement & Interview Takeaway

Interview Answer: Cannot assign requested address happens when a client exhausts its ephemeral port pool because closed TCP connections stay locked in the 60-second TIME_WAIT state. To fix this in production, never open raw HTTP connections per request; always use HTTP Keep-Alive connection pools to reuse established TCP sockets.

---

📺 Video Explanation