Download websites that use Tengine

About Tengine

Most forks of a widely deployed web server keep the identifier of the project they came from and vanish into its totals. Tengine changed the string it reports about itself, and that change is the only reason a measurement like this one exists at all. Everything that follows rests on operators leaving the identifier in place, which is a decision they can reverse in a single line of configuration.

The country column is the part that surprises people. The server was built inside a Chinese commerce operation and is maintained under Alibaba / Taobao, yet United States holds 359,540 of the sites where we see it and the suffix mix leans toward .com. The explanation is not that adoption migrated. Country in this dataset follows hosting and registration, and a great deal of internationally facing commerce is served from infrastructure registered outside the operator home market, on generic domain suffixes, through providers with a global footprint. Treat the column as a map of where traffic terminates rather than a map of who writes the configuration.

Presence itself is a strong qualifier, and that is the practical reason to care. The additions this fork makes over its upstream project, from health checking across upstream pools to dynamic module loading and request filtering, matter only above a certain level of traffic and only to a team that operates its own edge. Shared hosting does not default to it. A domain answering with this identifier is very likely run by an organization with an infrastructure team, own load balancing and a reason to have evaluated alternatives, which is exactly the profile that vendors of delivery networks, security filtering and observability tooling spend money to find.

The limits are the usual ones for server detection, and they bite harder here. Identification depends on a response header that operators frequently strip, rewrite or replace with a brand of their own. Anything behind a content delivery network usually presents the network identity instead of the origin. Downstream distributions rebuild the fork and rename it again. The result is an undercount that is not randomly distributed, because the operators most capable of running this software are the operators most likely to hide what they run. We also do not get a dependable release string from it, so the record here is presence rather than version, and a presence record should not be read as a version inventory.

Released under BSD terms in 2011, the project has no license registry and no vendor with a customer list to disclose, so observation from outside is the only way to size it. A per-domain lookup will tell you what one host answers with today. The downloadable list tells you which organizations, in which countries, under which suffixes, are running an edge they built themselves, and whether that population grows or shrinks between snapshots. The first question is a diagnostic. The second one is a market.