sandbox-benchmark

Overview ¶

Command sandbox-benchmark stress-tests sandbox creation latency by running a series of concurrency levels, launching N sandboxes at each level and measuring create API response time and time-to-running.

Each concurrency level is repeated R times (default 1). The summary reports mean±std across repeats for each percentile.

Usage:

export TENSORLAKE_API_KEY=<your-api-key>
go run ./examples/sandbox-benchmark -con 1,10,100
go run ./examples/sandbox-benchmark -repeat 5 -con 1,10,100,1000

Flags:

-con       Comma-separated concurrency levels (default: 1,10,100)
-repeat    Number of times to repeat each concurrency level (default: 1)
-timeout   Sandbox timeout in seconds (default: 120)
-poll      Poll interval for status checks (default: 100ms)
-max-wait  Maximum wait time per sandbox before giving up (default: 120s)