[Kst] 2.0.8

[Barth Netterfield]

Ben,

I have pushed a hack.  See if it improves the situation.  If it does, I can 
try to clean it up a bit.

cbn

On February 22, 2014 1:14:18 PM Barth Netterfield wrote:

> The fact that the problems happen with the configuration, below, and not on
> my system re-enforces my belief that this is file system race-condition
> related.
> 
> kst decides that a file has been re-written/replaced and needs to be re-read
> from the beginning when it has shrunk.  Apparently, in the configuration
> below, if a file is being written to, it can appear smaller than it was. 
> As far as I know, this can not happen in Linux.
> 
> As a hack (as opposed to a fix, which might require re-writing the file
> system I suspect) we could tell the data vector to delay reading a few
> times if the data source reports back with a shrunk file - and not try a
> re-read until it has failed a few times.  This will (dramatically?) reduce
> the rate of failures but since the problems are unpredictable, it won't
> formally fully fix the problem.
> 
> I will commit a hack shortly.
> 
> The second problem (reading zeros) is harder to fix (but probably caused by
> the same thing).  I will think about it.
> 
> cbn
> 
> On February 22, 2014 12:22:47 PM Ben Lewis wrote:
> > I have a USB memory stick plugged into the PC where the data is generated.
> > Data is collected to a RAM buffer and then written to a CSV file on the
> > memory stick. The memory stick has Windows file sharing enabled so that it
> > is accessible over the local network.
> > Kst runs on a different PC. The shared drive (memory stick) on the remote
> > PC is mapped to a local drive. Kst then reads the CSV file as if it were
> > a local file (with update type set to "time interval") The connection
> > between the two PCs is either via a LAN cable or an Ad-hoc WiFi
> > connection. The problem exists under both cases.
> > 
> > Remote System (where CSV data file is generated)
> > ----------------------------------------------------------------
> > OS: WindowsXP embedded (32bit)
> > File System: Data is written to a USB memory stick, formatted with NTFS
> > 
> > Data Accumulation Rate:
> >       fields/row: 5
> >       characters per field: 7
> >       bytes per row: 41
> >       rows per second: 1500
> > 
> > I've attached the first 100 lines of a data file.
> > 
> > Local System (where Kst is run)
> > ----------------------------------------
> > OS: Windows7 (64bit)
> > File System: Remote NTFS drive is mapped to local drive
> > Kst: 64bit build
> > 
> > >> *   "Out of Memory" error
> > >> http://kde.6490.n7.nabble.com/Out-of-memory-error-td1555215.html The
> > >> error
> > >> message has been improved but the message still appears when it
> > >> shouldn't.
> > > 
> > > I also can't reproduce this.  Can you give me details (OS, file size)? 
> > > Is
> > > the data being updated real time during the read?  At what rate?
> > 
> > Same as above
> > 
> > >>    * I sometimes get snippets of data missing in a live plot. If I
> > >>    restart
> > >> 
> > >> Kst and reload the data there are no missing bits. This seems to happen
> > >> when there is a large amount of network traffic (my data file is not on
> > >> a
> > >> local disk). This is hard to reproduce so it's probably not worth
> > >> worrying
> > >> about at this stage.
> > > 
> > > Yes.... are you using smb or nfs?  Is it at all correlated with the
> > > first
> > > bug?
> > 
> > I'm using Window file sharing.
> > It could very well be related to the first bug but I have no way of
> > telling.

-- 
Barth Netterfield
University of Toronto
416-845-0946