On Freitag, 8. April 2022 23:56:15 CEST Milian Wolff wrote:

<snip>

> Which in turn points at the code that does cache cleanup in
> `debuginfod_query_server`. I now used `rr` to record such a bogus run and I
> clearly see that `(time(NULL) - st.st_mtime <= cache_miss)` is false and it
> goes into the unlink case.
> 
> I'll try to debug this further now - I definitely do not wait 600s inbetween
> these runs here...

I compiled elfutils with debug output, and here's what I can see when I run 
`debuginfod-find`:

time(NULL) = 1649455510
st.st_mtime = 1649448650
delta = 6860
cache_miss = 600

The longer I wait, the bigger the delta becomes - i.e. for every second I 
wait, the `delta` increases by one.

And I think I know where this comes from:

```
# first we stat the target cache path
if (stat(target_cache_path, &st) == 0
  {

  # then we pass _the same st_ to
  debuginfod_config_cache(cache_miss_path, cache_miss_default_s, &st)

  # which internally will do 
  stat(config_path, st)

  # then we check the time delta
  time(NULL) - st.st_mtime <= cache_miss
```

I.e. when we check the time delta, we only take the time stamp of the 
`config_path` into account - the time stamp of the `target_cache_path` is 
ignored.

I mount my filesystems with relatime (old advise for ssd's, probably not 
relevant anymore?). I guess that's the issue then?

Can we change the above code to store the `st.st_mtime` for 
`target_cache_path` and use that for comparison purposes? That fixes the issue 
for my case. If this is acceptable, I'll provide a patch.

Thanks
-- 
Milian Wolff
mail@milianw.de
http://milianw.de