Running cold email infrastructure.

Sending domains started appearing on SURBL before they had sent anything. The cause was not the copy, not the domains, and not the volume. It was the DNS cluster they shared.

SURBL evaluates the domains inside the message body, not the sending domain. One listed link affects every message carrying it. So we removed links from cold email, and each sending domain now serves its own page.

Same list, same week, three sending setups, placement measured on our own seed inboxes. The third-party relay reached the inbox 16.8 percent of the time. The rented pool reached 55.6. Our own cluster beat both.

Warmup, domain age and sending history all accumulate against an IP address, and whoever owns that address owns the result. We hold our own IPv4 space across three isolated ranges, so changing host does not change the addresses.

Fleet capacity is a distributed-state problem: hundreds of domains whose status can be stale, missing, or wrong at read time. The design that keeps the number honest: explicit states, unknown never counts as clean, every rate ships its denominator, and sending halts when state cannot be read.

When one entity gets popular, one database partition takes the load meant for a fleet, and caching alone cannot save it. The clearest public answer is Bo Ingram's work at Discord: shape the traffic in a service you control, then change the engine.

Every product that wins eventually outgrows one Postgres box, and every path out is expensive: NewSQL migration, middleware, extensions, or sharding logic smeared through the codebase. The standout answer is Sammy Steele's work at Figma: shard logically first, route in one proxy, rehearse until the irreversible step is seconds long.