Re: kernel 2.6.25-rc7 highly unstable on high load

[Eric Dumazet <dada1@cosmosbay.com>]

Denys Fedoryshchenko a écrit :
> It looks like probably i did mistake while applying it. Very strange, cause i 
> rechecked it, and i did make clean before building system image.
> I will make sure on vanilla code that everything is fine.
> 
> Btw, i see permanently 4 "delayed"(?) records.
> 
> [ 6261.050827] dstgc [0] lo     09E9FED5        09E9FED5        
> 80000000        5       29      0       8F323353        0       0       
> 0       00      -1  009E9FED5
> [ 6261.050834] dstgc [1] lo     09E9FED5        09E9FED5        
> 80000000        37      358     0       FC9B62D4        0       0       
> 0       00      -1  009E9FED5
> [ 6261.050838] dstgc [2] lo     09E9FED5        09E9FED5        
> 80000000        4       8       0       7048DF48        0       0       
> 0       00      -1  009E9FED5
> [ 6261.050842] dstgc [3] lo     09E9FED5        09E9FED5        
> 80000000        6       5       0       6924E846        0       0       
> 0       00      -1  009E9FED5
> [ 6261.050851] dst_total: 129123 delayed: 4 work_perf: 0 expires: 68000 
> elapsed: 27 us
> 
> Especially interesting one with refcnt 37 and dst.__use 358. I check for 
> around 5 minutes - there is no traffic at all. This record staying for long 
> time and there is no activity on it over "tcpdump -ni any host 
> 212.98.155.252" , and it is not flushing it over ip route flush cache.
> 

This route with refcnt 37 has Source IP = FC9B62D4, which is 212.98.155.252

Check that you dont have some tcp sessions with this remote address :

ss dst 212.98.155.252

As long as you have no activity on it, refcnt wont change.