status of seekwater

status of seekwater

[Christoph Hellwig]

Has any tried to use seekwatcher lately?  For one the report on
oss.oracle.com seems to be gone.  I found a git repo that seems
to have the latest codebase from hg and two little fixes here:

https://github.com/trofi/seekwatcher

But even that doesn't seem to work in modern environments.  After
a build with a few warnings actually trying to run it creates doesn't
seem to actually work well:


$ /usr/local/bin/seekwatcher -o sda2.png -t sda2
using tracefile ./sda2
Traceback (most recent call last):
  File "/usr/local/bin/seekwatcher", line 983, in <module>
    run = run_blkparse(x)
  File "/usr/local/bin/seekwatcher", line 614, in run_blkparse
    loaddata(fh, rd)
  File "/usr/local/bin/seekwatcher", line 130, in loaddata
    tags, options)
  File "rundata.pyx", line 234, in seekwatcher.rundata.rundata.load_data (seekwatcher/rundata.c:3936)
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()

With not output produces. 

Re: status of seekwater

[Souvik Banerjee]

I had the same problem with seekwatcher when I tried to use it. I
ended up writing my own tool to visualize blktrace files which you can
find here: https://github.com/souvik1997/blkview

It’s not exactly the same but it might be helpful.

Thanks,

Souvik

On 4/14/19, Christoph Hellwig <hch@lst.de> wrote:
> Has any tried to use seekwatcher lately?  For one the report on
> oss.oracle.com seems to be gone.  I found a git repo that seems
> to have the latest codebase from hg and two little fixes here:
>
> https://github.com/trofi/seekwatcher
>
> But even that doesn't seem to work in modern environments.  After
> a build with a few warnings actually trying to run it creates doesn't
> seem to actually work well:
>
>
> $ /usr/local/bin/seekwatcher -o sda2.png -t sda2
> using tracefile ./sda2
> Traceback (most recent call last):
>   File "/usr/local/bin/seekwatcher", line 983, in <module>
>     run = run_blkparse(x)
>   File "/usr/local/bin/seekwatcher", line 614, in run_blkparse
>     loaddata(fh, rd)
>   File "/usr/local/bin/seekwatcher", line 130, in loaddata
>     tags, options)
>   File "rundata.pyx", line 234, in seekwatcher.rundata.rundata.load_data
> (seekwatcher/rundata.c:3936)
> ValueError: The truth value of an array with more than one element is
> ambiguous. Use a.any() or a.all()
>
> With not output produces.
>

Re: status of seekwater

[Chris Mason]
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Re: status of seekwater

[Christoph Hellwig]

On Sun, Apr 14, 2019 at 07:43:17PM +0000, Chris Mason wrote:
> I replaced seekwatcher with iowatcher a while ago, and convinced Jens to 
> stuff it into blktrace git.  It's dramatically faster (C instead of 
> python), and the command line is slightly different.  But, all the 
> features should be there.

Ok.  Can we add a note to the blktrace repo that this replaces seekwatcher
in a way that can be easily found?  Do we still have friends at Oracle
that could add such a note to the old seekwatcher page at:

    https://oss.oracle.com/~mason/seekwatcher/

Re: status of seekwater

[Bryan Gurney]

On Sun, Apr 14, 2019 at 3:43 PM Chris Mason <clm@fb.com> wrote:
>
> On 14 Apr 2019, at 7:46, Christoph Hellwig wrote:
>
> > Has any tried to use seekwatcher lately?  For one the report on
> > oss.oracle.com seems to be gone.  I found a git repo that seems
> > to have the latest codebase from hg and two little fixes here:
> >
> > https://github.com/trofi/seekwatcher
> >
> > But even that doesn't seem to work in modern environments.  After
> > a build with a few warnings actually trying to run it creates doesn't
> > seem to actually work well:
>
> I replaced seekwatcher with iowatcher a while ago, and convinced Jens to
> stuff it into blktrace git.  It's dramatically faster (C instead of
> python), and the command line is slightly different.  But, all the
> features should be there.
>

Hi Chris,

I use seekwatcher and iowatcher to visualize the output of very long
blktrace runs.  One of the examples that led to a patch was "xfs: skip
online discard during eofblocks trims" (unfortunately, I missed my
chance to get a "Reported-by" tag on that patch).  In that specific
case, seekwatcher was rendering the discards as writes (at least I
believe so; under seekwatcher, "green" was used for writes, correct?)

I ran another blktrace, and decided to start "iowatcher ...
--movie=rect", and I rediscovered a few "rough edges" on iowatcher:

1. I don't believe that iowatcher renders discards.  It would be nice
if it did render discards, in a way that could be distinguishable from
reads and writes, now that it is one of the "RWBS" operations in
blktrace / blkparse.

2. I'm having issues with the time scaling, where all of the data
points seem to end up at the "end" of the video.  I forgot what
combination of "--rolling" (rolling average in seconds) and/or
"--xzoom" parameters are required to re-scale the graph, but it
becomes a trial-and-error process, since the user needs to render the
video to see if any corrections need to be made.  (Thankfully,
rendering doesn't take over 12 hours for a 16 GiB trace anymore.)

Under seekwatcher, this set of parameters would work:

seekwatcher -t sdb1.blktrace.0 -o outfile.ogg --movie
--movie-lengthH0 --rolling-avg=1

3. I miss seekwatcher's "time scale" graph on the bottom, which would
display the "seconds" counter on an X axis of the "MB/s" and "seeks/s"
graphs.

If there's any way I can help, with examples to recreate, let me know.


Thanks,

Bryan