Hi Florin,
Hi Rajith,
It shouldn't be the pool expansion case, I have
8341f76fd1cd4351961cd8161cfed2814fc55103.
Moreover, in this case _e would be different from
&load_balance_pool[3604]. I've
found some of those expansions (in other places), in those cases a pointer
to the element has a different
Classify tables uses their own heaps.
> #3 0x76bdb6ca in vnet_classify_new_table (cm=0x77db9460
> , mask=0x7fffb9d52360 "", nbuckets=2,
> memory_size=1024,
> skip_n_vectors=1, match_n_vectors=2) at
> /home/Immaculate/Oct12/vpp/src/vnet/classify/vnet_classify.c:194
You asked for
Hello Mates,
I’m trying to create 50k classify tables with 6GB heap memory and 8GM
docker RAM but the vpp crashes after the creation of 32k classify tables.
Even though there are unused memory left in both heap and ram the vpp
crashes due to out-of -memory.
Is there any reason for it, if it
HI Stanislav,
My guess is you don't have the commit below.
commit 8341f76fd1cd4351961cd8161cfed2814fc55103
Author: Dave Barach
Date: Wed Jun 3 08:05:15 2020 -0400
fib: add barrier sync, pool/vector expand cases
load_balance_alloc_i(...) is not thread safe when the
load_balance_po
Hi Stanislav,
The only thing I can think of is that main thread grows the pool, or the pool’s
bitmap, without a worker barrier while the worker that asserts is trying to
access it. Is main thread busy doing something (e.g., adding routes/interfaces)
when the assert happens?
Regards,
Florin
Hi Florin,
I wasn't aware of those helper functions, thanks! But yeah, it also returns
0 (sorry, but there's the trace of another crash)
Thread 3 "vpp_wk_0" received signal SIGABRT, Aborted.
[Switching to Thread 0x7f9cc0f6a700 (LWP 3546)]
__GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux
Hi all,
just a note that VPP 21.10 RC2 images are available from the packagecloud repo.
Only the critical bugfixes can go now into the stable/2110 branch.
The release date is on the 27 October, according to the release plan [0].
thanks a lot!
--a /* your friendly 21.10 release manager */
[0]
Hi Stanislav,
Just to make sure the gdb macro is okay, could you run from gdb: pifi(pool,
index)? The function is defined in gdb_funcs.c.
Regards,
Florin
> On Oct 13, 2021, at 11:30 AM, Stanislav Zaikin wrote:
>
> Hello folks,
>
> I'm facing a strange issue with 2 worker threads. Sometimes
Hello folks,
I'm facing a strange issue with 2 worker threads. Sometimes I get a crash
either in "ip6-lookup" or "mpls-lookup" nodes. They happen with assert in
the *pool_elt_at_index* macro and always inside the "*load_balance_get*"
function. But the load_balance dpo looks perfectly good, I mean
Yes, mpls routes and ip routes that push label are not MP safe. I fixed
this in our node that programs these routes. Fix is taking the barrier lock
before programming such routes and releasing it once done.
-Rajith
On Wed, 13 Oct 2021 at 1:12 PM, Stanislav Zaikin wrote:
> Hi Rajith,
>
> Did you
Hi Rajith,
Did you find the root cause? I'm facing the same problem, the load_balance
element in the pool seems to be good. And event a helper from gdbinit says
that the element isn't free:
#0 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:51
51 ../sysdeps/unix/sysv/linux/rai
11 matches
Mail list logo
