pbxt-discuss team mailing list archive

[Paul McCullagh <paul.mccullagh@xxxxxxxxx>]

Hi Erkan,

Yes. Strange pause. But I cannot see what the engine is doing in that  
time.

There must be activity in some missing statistics, for example,  
record, row or index file writing.

Unfortunately, when displaying all statistics, the output is very  
wide. What xtstat needs is a feature which logs all statistics but  
just displays what you select.

Then, if you see a problem that does not show up in the displayed  
statistics you can still go back and check the others at the time.

On Nov 30, 2010, at 3:05 AM, erkan yanar wrote:

> Moin Paul
>
> On Mon, Nov 29, 2010 at 10:23:38AM +0100, Paul McCullagh wrote:
> > The problem here is the index cache is full, and the thread which is
> > flushing the index is hanging in sync (i.e. flush to disk). The rest
> > of the system is hanging because it needs a clean page in the index.
> >
> > I did quite a bit of work to improve this in 1.1 and 1.5.
> >
> > But in general, if the index cache is really large, it need to be
> > flushed more regularly to ensure that too many changes do not
> > accumulate. Checkpointing should do this, but there is a bug in 1.0.
> >
> > You checkpoint frequency is 128M, xtstat says to-chkpt = 1049M. This
> > means the checkpoint has not occurred as often as he should.
> >
> > This problem should be fixed in 1.1.
>
> While 1.5.02 Beta handles IO a lot better than 1.0.11-7, you may be
> interested in that checkpointing issue:
> pbxt_checkpoint_frequency    = 67108864
>
>           time  time  xact  xact  xact  xact  stat  stat  ilog   
> ilog  ilog  ilog     ilog  xlog  xlog  xlog  xlog  xlog  xlog  xlog   
> xlog    to    to    to sweep   row   row   row   row
>           curr  msec commt rollb waits dirty  read write    in   out  
> rd.ms wr.ms syncs/ms    in   out rd.ms wr.ms syncs sy.ms  hits  miss  
> chkpt write sweep waits   sel   ins   upd   del
> 101130 03:45:13  5001  198t     0     0 10.9t     0  198t  320M      
> 0    83     0      0/0     0 21.1M     0    12     9  3576 69.6t      
> 0  813M 9833K     0    13     0  198t     0     0
> 101130 03:45:18  5001  9501     0     0  5148     0  9501     0      
> 0     0     0      0/0     0     0     0     0     0  5001     0      
> 0  813M 9833K     0     1     0  9501     0     0
> 101130 03:45:23  5001     0     0     0  5148     0     0     0      
> 0     0     0      0/0     0     0     0     0     0  5001     0      
> 0  813M 9833K     0     0     0     0     0     0
> 101130 03:45:28  5001 80.1t     0     0 13.8t     0 80.1t     0      
> 0     0     0      0/0     0 8192K     0     4     1  4993     1      
> 0  814M 17.6M  115M     5     0 80.1t     0     0
> 101130 03:45:33  5001  179t     0     0 10.3t     0  179t     0      
> 0     0     0      0/0 1244K 19.7M     0     8    20  4395  119t     
> 39  837M 6772K     0    13     0  179t     0     0
> 101130 03:45:38  5001 12.7t     0     0  9084     0 12.7t     0      
> 0     0     0      0/0     0     0     0     0     0  5001  5507      
> 0  837M 3756K     0     1     0 12.7t     0     0
> 101130 03:45:43  5001 78.4t     0     0  8907     0 78.4t     0      
> 0     0     0      0/0     0 8192K     0     3     1  4994 18.1t      
> 0  840M 8192K     0     5     0 78.4t     0     0
> 101130 03:45:48  5001  156t     0     0  4410     0  156t     0      
> 0     0     0      0/0     0 16.0M     0     8     2  4985 67.5t      
> 0  856M 8192K     0    11     0  156t     0     0
> 101130 03:45:53  5001 78.9t     0     0  7213     0 78.9t     0      
> 0     0     0      0/0     0 8192K     0     6     1  4991     1      
> 0  864M 16.0M     0     7     0 78.9t     0     0
>
>
> Best Regards
> Erkan
>
> --
> über den grenzen muß die freiheit wohl wolkenlos sein