I blamed HTTP/2 multiplexing. Then I counted the connections.
Every request succeeded. The log stayed 33,000 entries behind. The mechanism I blamed was real, present, and not the problem.
Engineering notes
Blog
Problems found while running a Certificate Transparency streaming server against every trusted log, and what the fixes turned out to be. Measurements included, wrong turns included. RSS
The CT logs that serve tiles but publish no checkpoint
Three logs, zero errors, 279,000 entries behind. The fast way to read them was sitting right there, behind a missing file.
358 MB resident, 52 MB in use: where the rest went
A service that looked like it was leaking. It wasn't, and the fix was five settings the allocator had been running without.