<font size=2 face="sans-serif">Thanks.</font>
<br>
<br><font size=2 face="sans-serif">While we're at it, can we somehow save
the index to a separate file, and allow a reload when re-opening the trace?
That would speed up opening considerably...</font>
<br>
<br><font size=2 face="sans-serif">I am about to attempt this, and I'll
submit a patch if it works for me.</font>
<br>
<br><font size=2 color=#000080 face="sans-serif">Amit Margalit</font>
<br><font size=2 color=#808000 face="sans-serif">IBM XIV </font><font size=2 face="sans-serif">-
<i>Storage Reinvented</i></font>
<br><font size=2 face="sans-serif">XIV-NAS Development Team</font>
<br><font size=2 face="sans-serif">Tel. 03</font><font size=2 face="Arial">-689-7774</font>
<br><font size=2 face="Arial">Fax. 03-689-7230</font>
<br>
<br>
<br>
<br><font size=1 color=#5f5f5f face="sans-serif">From:
</font><font size=1 face="sans-serif">Julien Desfossez <jdesfossez@efficios.com></font>
<br><font size=1 color=#5f5f5f face="sans-serif">To:
</font><font size=1 face="sans-serif">Amit Margalit/Israel/IBM@IBMIL</font>
<br><font size=1 color=#5f5f5f face="sans-serif">Cc:
</font><font size=1 face="sans-serif">Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
lttng-dev@lists.lttng.org</font>
<br><font size=1 color=#5f5f5f face="sans-serif">Date:
</font><font size=1 face="sans-serif">07/25/2013 07:37 PM</font>
<br><font size=1 color=#5f5f5f face="sans-serif">Subject:
</font><font size=1 face="sans-serif">Re: [lttng-dev]
Getting SIGBUS in babeltrace - occasionally</font>
<br>
<hr noshade>
<br>
<br>
<br><tt><font size=2>> Is there a way to determine the "latest
timestamp that is safe to access" ?<br>
I guess with the switch-timers enabled on all channels (including<br>
metadata) you can estimate what is the maximum timestamp you can read.<br>
<br>
For example, if your switch-timer is set to 1 second, you can check the<br>
current time, wait a 1 second grace period and read the trace up to that<br>
point, but I wouldn't consider that a safe or clean method.<br>
Also, you need to reopen the trace every time since the index is done<br>
only when opening the trace and metadata might get appended during the<br>
trace.<br>
<br>
I don't know exactly what your use-case is, but the live reading seems<br>
to be what you are looking for. We will post more details about it soon.<br>
<br>
Thanks,<br>
<br>
Julien<br>
<br>
<br>
> <br>
> Currently, I am using get_timestamp_end() on the trace handle and
then I<br>
> try to read the events at that point, and if I fail, I simply try<br>
> earlier times until I succeed in reading an event.<br>
> <br>
> Thanks,<br>
> <br>
> Amit Margalit<br>
> IBM XIV - /Storage Reinvented/<br>
> XIV-NAS Development Team<br>
> Tel. 03-689-7774<br>
> Fax. 03-689-7230<br>
> <br>
> <br>
> <br>
> From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com><br>
> To: Amit Margalit/Israel/IBM@IBMIL<br>
> Cc: lttng-dev@lists.lttng.org<br>
> Date: 07/24/2013 07:14 PM<br>
> Subject: Re: [lttng-dev] Getting SIGBUS
in babeltrace - occasionally<br>
> ------------------------------------------------------------------------<br>
> <br>
> <br>
> <br>
> * Amit Margalit (AMITM@il.ibm.com) wrote:<br>
>> Hi,<br>
>><br>
>> Access beyond end of file seems to be the issue IIUC.<br>
>><br>
>> Let me try to give as much context as I can and answer your questions
one<br>
>> by one:<br>
>><br>
>> Versions<br>
>><br>
>> lttng-ust 2.2.0, lttng-tools 2.2.0, babeltrace - I used git hash<br>
>> cc26a15ac355d886bb507955fdaec9b218024713 with a patch on babeltrace.i.in.<br>
>><br>
>> [I cannot supply the patch as this requires my employer's (IBM)<br>
> approval -<br>
>> I am in the process of obtaining a permanent contribution approval,
but<br>
>> until then...]<br>
>><br>
>> However, I can tell you exactly what it does:<br>
>> 1. Adds a typedef for 'ssize_t'<br>
>> 2. Added a call to _bt_ctf_get_decl_from_def() around the argument<br>
>> supplied to _bt_ctf_get_int_len()<br>
>> 3. Changed all use of {} in string ".format()" calls
to add numbers (e.g.<br>
>> {0} instead of {}) - as this doesn't work for Python 2.6.<br>
>><br>
>> I am waiting for the contribution approval and then I'll submit
a patch.<br>
>><br>
>> To provide a trace sample, I'll need a special approval, which
could take<br>
>> very long.<br>
>><br>
>> System Setup<br>
>><br>
>> Processor: Xeon-family 6-cores (12 if you count HT)<br>
>> RAM: 24GB of RAM, page size is 4KB.<br>
>> OS: Linux 2.6.32.12-211_11_4_0 based on an MCP (IBM's SuSE derivative)<br>
>> build.<br>
>><br>
>> Tracing Setup<br>
>><br>
>> Same system used to generate, collect and read the traces.<br>
>> 100% user-space.<br>
>><br>
>> Triggering<br>
>><br>
>> This is not readily reproducible. Some context:<br>
>><br>
>> We have an in-house trace mechanism including a code instrumentor
and a<br>
>> curses-based viewer (actually urwid+python).<br>
>> My current project is to embed LTTng logging and trace capabilites
into<br>
>> several open-source projects, because we couldn't use our in-house
tracer<br>
>> without having GPL issues.<br>
>> I have written a back-end that enables our in-house viewer to
display<br>
>> Lttng traces, by linking with libbabeltrace (allowed by its LGPL
license).<br>
>><br>
>> As far as I can tell SIGBUS happens to me when I was doing a search.
This<br>
>> literally reads the events in sequence and tries to compare some
fields<br>
>> against some criteria.<br>
>><br>
>> And here's the main reason why I think this is happening - I was
adding a<br>
>> trace (using add_traces_recursive(trace_dir,'ctf')) while the
session is<br>
>> active (i.e. not stopped).<br>
>><br>
>> I can see 2 problematic scenarios here:<br>
>><br>
>> Scenario 1<br>
>><br>
>> When I add the trace to the context babeltrace performs mmap()
according<br>
>> to the size of the file(s) at that moment.<br>
>> The files keep growing.<br>
>> When I try to access new packets they may lie beyond the end of
the<br>
>> mmap'ed region. I think this should lead to SIGSEGV and not to
SIGBUS.<br>
> <br>
> babeltrace only mmap packet by packet, never an entire file.<br>
> <br>
>><br>
>> Scenario 2<br>
>><br>
>> Due to buffering issues, I could be having a race with the session<br>
> daemon,<br>
>> where my code reads the beginning of a packet that was already
written to<br>
>> the file, and try to access the contents, which were not yet written.<br>
>> Again - could happen around the end of the file, with the session
still<br>
>> active.<br>
> <br>
> I think the reason is close to this scenario: I think that babeltrace<br>
> try to read a packet as it is being written to disk. So it gets the<br>
> packet header (including the expected size of the packet), and attempts<br>
> to read beyond the end of the partially-copied packet. Live streaming<br>
> feature (we're working on this for the next release of lttng) will<br>
> handle thi use-cases.<br>
> <br>
> Thanks,<br>
> <br>
> Mathieu<br>
> <br>
> <br>
>><br>
>> Amit<br>
>><br>
>> Amit Margalit<br>
>> IBM XIV - Storage Reinvented<br>
>> XIV-NAS Development Team<br>
>> Tel. 03-689-7774<br>
>> Fax. 03-689-7230<br>
>><br>
>><br>
>><br>
>> From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com><br>
>> To: Amit Margalit/Israel/IBM@IBMIL<br>
>> Cc: lttng-dev@lists.lttng.org<br>
>> Date: 07/24/2013 04:20 PM<br>
>> Subject: Re: [lttng-dev] Getting SIGBUS
in babeltrace -<br>
>> occasionally<br>
>><br>
>><br>
>><br>
>> * Mathieu Desnoyers (mathieu.desnoyers@efficios.com) wrote:<br>
>> > * Amit Margalit (AMITM@il.ibm.com) wrote:<br>
>> > > Hi,<br>
>> > ><br>
>> > > I am getting occasional SIGBUS inside _aligned_integer_read()
in<br>
>> > > ./formats/ctf/types/integer.c<br>
>> > ><br>
>> > > Here is the backtrace, maybe someone could take a look
and suggest<br>
>> > > something:<br>
>> ><br>
>> > It looks like you're hitting the internal checks for overflow
on the<br>
>> > buffer boundaries.<br>
>><br>
>> Please forget about my previous email, I'm utterly wrong. That
should<br>
>> teach me to reply before my first morning coffee ;)<br>
>><br>
>> We indeed have internal checks for overflows in babeltrace, but
those<br>
>> are not triggering SIGBUS ever, they return failure gracefully.<br>
>><br>
>> What seems to happen in your case is documented here:<br>
>><br>
>> mmap(2) manpage:<br>
>><br>
>> SIGBUS Attempted access to a portion
of the buffer that does not<br>
>> corre$B!>(B<br>
>> spond to
the file (for example, beyond the end of the<br>
>> file,<br>
>> including the
case where another process has truncated<br>
>> the<br>
>> file).<br>
>><br>
>> I'm curious to learn which versions of babeltrace/ust/modules
you are<br>
>> using, if you modified them (and how), and if you can get us a
sample<br>
>> trace that triggers the issue.<br>
>><br>
>> Also, a bit of context about your setup: is this 32-bit or 64-bit<br>
>> user-space, which OS.<br>
>><br>
>> Hrm.<br>
>><br>
>> By any chance, is it possible that you record your trace on a
platform<br>
>> with 4kB pages, and run babeltrace on it on a platform having
64kB<br>
>> pages? Everything points me to include/babeltrace/mmap-align.h<br>
>> mmap_align_addr(). We might be mmaping beyond the end of file
in this<br>
>> case.<br>
>><br>
>> Thanks!<br>
>><br>
>> Mathieu<br>
>><br>
>> ><br>
>> > On which babeltrace version can you reproduce it ? Did you
do any<br>
>> > modification to babeltrace on your own on top ?<br>
>> ><br>
>> > Are you getting this with a lttng-ust or lttng-modules trace
? If yes,<br>
>> > which version of the tool has generated the trace ?<br>
>> ><br>
>> > Can you provide a sample trace that triggers this issue ?<br>
>> ><br>
>> > Can you give us the detailed sequence of steps you use to
reproduce ?<br>
>> ><br>
>> > Thanks,<br>
>> ><br>
>> > Mathieu<br>
>> ><br>
>> > > #0 _aligned_integer_read (definition=0x954320,
ppos=0x106e028) at<br>
>> > > integer.c:81<br>
>> > > #1 ctf_integer_read (ppos=0x106e028, definition=0x954320)
at<br>
>> > > integer.c:224<br>
>> > > #2 0x00007ffff070e36e in generic_rw (definition=<optimized
out>,<br>
>> > > pos=0x106e028) at ../include/babeltrace/types.h:133<br>
>> > > #3 bt_struct_rw (ppos=0x106e028, definition=0x9542e0)
at struct.c:56<br>
>> > > #4 0x00007ffff13f7fda in generic_rw (definition=<optimized
out>,<br>
>> > > pos=0x106e028) at ../../include/babeltrace/types.h:133<br>
>> > > #5 ctf_packet_seek (stream_pos=0x106e028, index=<optimized
out>,<br>
>> > > whence=<optimized out>) at ctf.c:860<br>
>> > > #6 0x00007ffff070a3f1 in seek_file_stream_by_timestamp<br>
>> > > (cfs=cfs@entry=0x106cf90,<br>
>> timestamp=timestamp@entry=1374636821498972926)<br>
>> > > at iterator.c:141<br>
>> > > #7 0x00007ffff070aa96 in seek_ctf_trace_by_timestamp<br>
>> > > (stream_heap=0x176d310, timestamp=1374636821498972926,
tin=<optimized<br>
>> > > out>) at iterator.c:188<br>
>> > > #8 bt_iter_set_pos (iter=iter@entry=0xfdafa0,
iter_pos=0x2bf59f0) at<br>
>> > > iterator.c:439<br>
>> > > #9 0x00007ffff1624f01 in _wrap__bt_iter_set_pos
(self=<optimized<br>
>> out>,<br>
>> > > args=<optimized out>) at babeltrace_wrap.c:3805<br>
>> > > #10 0x00007ffff7b28c0c in call_function (oparg=<optimized
out>,<br>
>> > > pp_stack=<optimized out>) at Python/ceval.c:3679<br>
>> > > #11 PyEval_EvalFrameEx (f=0x1c243c0, throwflag=<optimized
out>) at<br>
>> > > Python/ceval.c:2370<br>
>> > > #12 0x00007ffff7b2e0d2 in PyEval_EvalCodeEx (co=0xa8e5d0,<br>
>> > > globals=<optimized out>, locals=<optimized
out>, args=0x2088cc8,<br>
>> > > argcount=0, kws=0x0, kwcount=0, defs=0x0, defcount=0,
closure=0x0)<br>
>> > > at Python/ceval.c:2942<br>
>> > > .<br>
>> > > .<br>
>> > > .<br>
>> > ><br>
>> > > Any suggestions are welcome.<br>
>> > ><br>
>> > > Amit Margalit<br>
>> > > IBM XIV - Storage Reinvented<br>
>> > > XIV-NAS Development Team<br>
>> > > Tel. 03-689-7774<br>
>> > > Fax. 03-689-7230<br>
>> > > _______________________________________________<br>
>> > > lttng-dev mailing list<br>
>> > > lttng-dev@lists.lttng.org<br>
>> > > </font></tt><a href="http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev"><tt><font size=2>http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev</font></tt></a><tt><font size=2><br>
>> ><br>
>> ><br>
>> > --<br>
>> > Mathieu Desnoyers<br>
>> > EfficiOS Inc.<br>
>> > </font></tt><a href=http://www.efficios.com/><tt><font size=2>http://www.efficios.com</font></tt></a><tt><font size=2>
<</font></tt><a href=http://www.efficios.com/><tt><font size=2>http://www.efficios.com/</font></tt></a><tt><font size=2>><br>
>><br>
>> --<br>
>> Mathieu Desnoyers<br>
>> EfficiOS Inc.<br>
>> </font></tt><a href=http://www.efficios.com/><tt><font size=2>http://www.efficios.com</font></tt></a><tt><font size=2>
<</font></tt><a href=http://www.efficios.com/><tt><font size=2>http://www.efficios.com/</font></tt></a><tt><font size=2>><br>
>><br>
>><br>
> <br>
> -- <br>
> Mathieu Desnoyers<br>
> EfficiOS Inc.<br>
> </font></tt><a href=http://www.efficios.com/><tt><font size=2>http://www.efficios.com</font></tt></a><tt><font size=2>
<</font></tt><a href=http://www.efficios.com/><tt><font size=2>http://www.efficios.com/</font></tt></a><tt><font size=2>><br>
> <br>
> <br>
> <br>
> <br>
> _______________________________________________<br>
> lttng-dev mailing list<br>
> lttng-dev@lists.lttng.org<br>
> </font></tt><a href="http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev"><tt><font size=2>http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev</font></tt></a><tt><font size=2><br>
> <br>
<br>
</font></tt>
<br>