On Fri, Sep 17, 2004 at 09:29:44AM +0200, Daniel Stenberg wrote:
> On Thu, 16 Sep 2004, Dan Fandrich wrote:
>
> >I just did an i386 uClibc build of curl and had some problems with tests
> >503 and 504. The first few tries, the test suite would hang on test 503; I
> >was able to abort the test once after which test 504 would immediately
> >fail. Eventually, I got a run that went through all the tests without
> >hanging, but test 503 failed. I noticed a few other test runs on various
> >architectures that also fail recently on 503 or 504:
>
> Do you still get problems if you run these days stand-alone, as a single
> test?

Test 503 always seems to run to completion on its own, but fails. Test 504
usually completes and succeeds, but sometimes hangs (for at least 15 min.)
I just caught a stack trace as it hung:

#0 0x400581a3 in socketcall () from /usr/i386-linux-uclibc/lib/libc.so.0
#1 0x4004a31e in recv () from /usr/i386-linux-uclibc/lib/libc.so.0
#2 0x40048a85 in dns_lookup () from /usr/i386-linux-uclibc/lib/libc.so.0
#3 0x40049413 in gethostbyname () from /usr/i386-linux-uclibc/lib/libc.so.0
#4 0x400248f5 in Curl_getaddrinfo (conn=0x805a000,
    hostname=0x8059f70 "localhost", port=55555, waitp=0xbfffe2d8)
    at ../../curl-7.12.2-20040916/lib/hostip4.c:356
#5 0x4000b456 in Curl_resolv (conn=0x805a000, hostname=0x8059f70 "localhost",
    port=55555, entry=0xbfffe394)
    at ../../curl-7.12.2-20040916/lib/hostip.c:428
#6 0x40018881 in CreateConnection (data=0x804d000, in_connect=0x8057fcc,
    addr=0xbfffeb88, async=0xbfffebcb "")
    at ../../curl-7.12.2-20040916/lib/url.c:3259
#7 0x40018b38 in Curl_connect (data=0x804d000, in_connect=0x8057fcc,
    asyncp=0xbfffebcb "") at ../../curl-7.12.2-20040916/lib/url.c:3407
#8 0x40022620 in curl_multi_perform (multi_handle=0x8058000,
    running_handles=0xbfffec14) at ../../curl-7.12.2-20040916/lib/multi.c:372
#9 0x0804873c in test (URL=0xbffff999 "http://127.0.0.1:8991/504")
    at ../../../curl-7.12.2-20040916/tests/libtest/lib504.c:49
#10 0x4003962e in __uClibc_main () from /usr/i386-linux-uclibc/lib/libc.so.0
#11 0x08048976 in _start ()

Timing out while resolving "localhost"??? Looks like a uClibc problem of
some sort and, indeed, it seems to open /etc/hosts but not read from it.
I just confirmed that with a "curl http://localhost", which sometimes
returned a "Couldn't resolve host" message and sometimes hung.

On the hunch that this was the same cause of the problems with test 503, I
edited the test script to set the proxy address to 127.0.0.1:8990 and sure
enough, the test now passes. It looks like tests 503, 504 and 509 are the
only ones that need to resolve "localhost". I suppose I should either fix my
uClibc setup or submit a patch to these tests ;-)

So, this wasn't a curl problem at all, but it's still an interesting
coincidence that these three tests (and test 63) are the *only* tests
currently showing some failures in the autobuild logs.

> >These are the NTLM with proxy tests that were having problems a year ago.
> >Has anything changed that might have caused them to sporadically fail?
>
> No, these are not NTLM with proxy tests. These are multi interface tests
> over a proxy. It looks as if there's a problem with the server or
> something...

Ok, I mis-extrapolated from a message to the list last year discussing
similar failures.

>>> Dan

-- 
http://www.MoveAnnouncer.com              The web change of address service
          Let webmasters know that your web site has moved