The Twenty-Hour Fuse
Two nights of debugging, one incident left to write about. The night of August 14th, a Claude Code session hung for an hour inside the VS Code extension, and I traced it to a per-token streaming stall — the previous two pieces, Two Clocks, Neither Lying and Coalescing the Stream , cover that hunt and the middleware that fixed it. The morning after, the same four-core box nearly went down under a…
The Twenty-Hour Fuse
On August 14th, a Claude Code session froze for an hour inside the VS Code extension, leading to a deeper investigation into the root cause. The following day, the same issue resurfaced, this time causing server instability under a high load average of 38.7. The problem had been brewing for twenty hours before the fuse finally blew at 11:09.
The machine's load numbers revealed a pattern: 11.08, 38.69, and 23.59 for one-minute, five-minute, and fifteen-minute intervals. While the load numbers were falling, the kernel's stall accounting for CPU, memory, and IO, known as PSI, showed memory usage at 10 percent and IO pressure at 45 percent. This marked the beginning of an IO storm, with processes parked in uninterruptible IO waits contributing to the load average.
The culprit was traced to a specific process: llama.cpp, consuming 3.9 GB of memory, seven Java services at 2.3 GB combined, and k3s at 1.1 GB. Additionally, 59 vscode-server processes were consuming 7.3 GB of memory, all of which was not typical behavior. The root cause was discovered to be the accumulation of processes from multiple servers and windows, with the decisive hour spent mapping the process tree to identify the issue.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
