zero-soak report

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Resilience check. We blackhole the backend's IP (iptables DROP on the OUTPUT chain) for the kill window, so the app must keep serving while its backend connection is dead (graceful degradation). After traffic is restored we assert the server process is still alive and that the data route works again. A missed health probe during the blackhole is expected (a worker can block on the dropped connection) and is NOT a failure — only an actual process death or a failure to recover counts.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Resilience check. We blackhole the backend's IP (iptables DROP on the OUTPUT chain) for the kill window, so the app must keep serving while its backend connection is dead (graceful degradation). After traffic is restored we assert the server process is still alive and that the data route works again. A missed health probe during the blackhole is expected (a worker can block on the dropped connection) and is NOT a failure — only an actual process death or a failure to recover counts.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Resilience check. We blackhole the backend's IP (iptables DROP on the OUTPUT chain) for the kill window, so the app must keep serving while its backend connection is dead (graceful degradation). After traffic is restored we assert the server process is still alive and that the data route works again. A missed health probe during the blackhole is expected (a worker can block on the dropped connection) and is NOT a failure — only an actual process death or a failure to recover counts.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Resilience check. We blackhole the backend's IP (iptables DROP on the OUTPUT chain) for the kill window, so the app must keep serving while its backend connection is dead (graceful degradation). After traffic is restored we assert the server process is still alive and that the data route works again. A missed health probe during the blackhole is expected (a worker can block on the dropped connection) and is NOT a failure — only an actual process death or a failure to recover counts.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Resilience check. We blackhole the backend's IP (iptables DROP on the OUTPUT chain) for the kill window, so the app must keep serving while its backend connection is dead (graceful degradation). After traffic is restored we assert the server process is still alive and that the data route works again. A missed health probe during the blackhole is expected (a worker can block on the dropped connection) and is NOT a failure — only an actual process death or a failure to recover counts.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.

What this checks
Memory-stability check. We sample the server's resident set size (VmRSS) every 10s across the soak window, take the median of the first third as the baseline, and compare it to the final sample. Growth is only called a leak if it exceeds BOTH the relative threshold (leak_rel_pct) and the absolute threshold (leak_abs_mib). A flat / no-sustained-growth plateau means no leak.