Latency regression with later versions

[penberg]

With latest version:

penberg@vonneumann workers-libsql % wrk --latency  https://hello-libsql.penberg.workers.dev/hello
Running 10s test @ https://hello-libsql.penberg.workers.dev/hello
  2 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   156.71ms   86.89ms 356.07ms   59.00%
    Req/Sec    28.91     16.09    60.00     57.58%
  Latency Distribution
     50%  173.80ms
     75%  236.54ms
     90%  264.57ms
     99%  356.07ms
  100 requests in 10.09s, 58.28KB read
Requests/sec:      9.91
Transfer/sec:      5.77KB

with 0.16.2:

penberg@vonneumann workers-libsql % wrk --latency  https://hello-libsql.penberg.workers.dev/hello
Running 10s test @ https://hello-libsql.penberg.workers.dev/hello
  2 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    52.98ms   11.26ms 162.18ms   77.93%
    Req/Sec    94.35     14.34   121.00     62.63%
  Latency Distribution
     50%   51.16ms
     75%   58.07ms
     90%   65.42ms
     99%   93.72ms
  1876 requests in 10.06s, 8.56MB read
  Non-2xx or 3xx responses: 1876
Requests/sec:    186.43
Transfer/sec:      0.85MB

[psarna]

for context: the main reason is the fact we switched to WebSocket by default, and that needs a round of handshakes before sending actual queries.

Also, care sharing the source code of hello-libsql, so that I can reproduce later? :innocent:

[psarna]

long fixed by going back to good ol' http!