[colug-432] Troubleshooting Suggestions: File Truncation Issue

[William E. T.]

Sorry my description wasn't clear.  The link created is a hard link; I
thought this is a fairly common pattern of writing data to disk for unix
systems.

The most concise way to precisely convey what's happening may be with some
trace exerts:

Expected:

open("main.pre.tsidx", O_RDWR|O_CREAT|O_EXCL|O_TRUNC, 0600) = 3

write(3, "\33[?1049h\33[?1h\33=\33[1;61r\33[?12;25h\33"..., 29549) = 29549

close(3)                                = 0

link("main.pre.tsidx", "main.tsidx")    = 0

unlink("main.pre.tsidx")                = 0

stat("main.tsidx", {st_mode=S_IFREG|0600, st_size=29549, ...}) = 0


So we write 29549 bytes to main.pre.tsidx so we would expect main.tsidx to
have a size of 29549 bytes and in this example it does (the st_size in the
stat).


If we iterate over that process thousands to millions of times, we'll see
the following happen:

open("main.pre.tsidx", O_RDWR|O_CREAT|O_EXCL|O_TRUNC, 0600) = 3

write(3, "\33[?1049h\33[?1h\33=\33[1;61r\33[?12;25h\33"..., 29549) = 29549

close(3)                                = 0

link("main.pre.tsidx", "main.tsidx")    = 0

unlink("main.pre.tsidx")                = 0

stat("main.tsidx", {st_mode=S_IFREG|0600, st_size=0, ...}) = 0


In this case, we would expect main.tsidx to have a size of 29549 bytes, but
it has a size of 0

The actual application is a proprietary application; its a niche product,
but its run by a number of companies in town.  The pastebin in the first
e-mail (http://pastebin.com/mFmZzYEL) has the a trace excerpt it from the
actual application.

The above examples are from tracing the C program I wrote to isolate this
problem (https://gist.github.com/w3ttr3y/167b349d2ab67e3aa9d2  it is ugly
and probably missing some checks to ensure things suceeded, so I'd only
trust it if you trace it and then verify everything suceeded in the trace);


The big difference between the proprietary application and my demo is the
real one uses explicit paths instead of relative; both files are in the
same directory though so crossing a partition boundary isn't an issue

I didn't explicitly answer your questions, question by question. I started
to, but I started thinking of all sorts of ambiguities so I hoped trace
excerpts would more concisely document what is happening.

I really appreciate you taking the time to try and help me,

--Bill
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.colug.net/pipermail/colug-432/attachments/20160110/96e9890b/attachment.html