http://qizmt.googlecode.com/svn/wiki/images/Qizmt_logo_small.png' alt='Qizmt logo (small)' />

Back to Wiki Main / Qizmt FAQ

Qizmt FAQ - Troubleshoot

When I try to open a text file in MR.DFS with "qizmt edit WordCount_Output.txt" I get an error.

Currently only support editing job files (the green ones) But the top of the file can be viewed with the head command.
e.g.
qizmt head WordCount_Output.txt
qizmt head WordCount_Output.txt 100 (to view 100 lines etc.)

Also, can pull the file out of mr.dfs and view it:

qizmt get WordCount_Output.txt \\<hostname>\D$\folder\WordCount_Output.txt && notepad \\<hostname>\D$\folder\WordCount_Output.txt

Killall does not work in production.

When there is a permission issue on some production clusters killall may not work. In this case can use an alternative way to killall distributed processes in the cluster:

By the way, to killall on production machine is qizmt killall proxy –f then issue qizmt health until it says 100% healthy

System.Exception:Insufficient resources for this job on cluster (ZBlock value file size > ZFILE_MAX_BYTES) (consider increasing sub process count)

Under IOSettings of a grouped mapreduce job can add a Setting to increase the number of intermediate blocks so that they are smaller and not too much tries to load at once per core:

e.g. <IOSettings> <Setting name="Subprocess_IntermediateBlockPrime" value="499" /> ... </IOSettings>

The default zblock count is 271, which is how many chunks per core in the cluster a file is broken into for the sort phase. Subprocess_IntermediateBlockPrime must be set to a prime number.

There is a qizmt command nearprime which can be used to get a prime number near some value, e.g.

D:>**qizmt nearprime** 500

500 is not prime

499 is nearest prime less than 500

503 is nearest prime greater than 500

Note also, regardless of increasing Subprocess_IntermediateBlockPrime count, all identical keys go into the same zblock. If this is the case, can have the key also include a random number between 0 and 100 so that all identical keys are copped into at least 100 different keys.

In the case that it’s too much going to a single key, here is an example of breaking it up with random number in key. But if it is not too much going into a single key, can just increase the Subprocess_IntermediateBlockPrime so that less keys get loaded at once into the reducers.

``` int,int,double . . .

recordset rKey = recordset.Prepare();
rKey.PutInt(random.Next(100));
.
.
.

} ]]> ```

System.Exception: mstring cannot accommodate your string operations. Please use C# string or StringBuilder instead.

If doing an extensive amount of appending, then declare a StringBuilder outside of the map() or reduce() and reuse it in-between each iteration.

Can just use ToString, no need to use mstring.

The exception is related to out of memory, so on your dev machine it will occur sooner, in production it will be able to grow to the max of byte[] for all mstrings in the current itteration. Can use qizmt ps to see available memory and any other running jobs that may be using memory.

It looks like my job is hanging. Is there a way to say what happening to it?

If there are errors happening, you can see them early with:

qizmt slaveloglist

can also clear slave logs with:

qizmt slavelogdelete

Error: Unable to check user logs on ‘SOMEHOST’

It looks like there is too much going tout to Qizmt_Log() this is only for debugging not for transporting gigabytes of data. Can fix by throttling to only let each reduce or mapper log up to a handful of Qizmt_Log() calls.

Could do something like: public int logfive = 5; public virtual void Map(ByteSlice line, MapOutput output) { . . . if(error_condition) { if(logfive > 0) { Qizmt_Log(someerror); } --logfive; } Or, you can configure qizmt to allow only a maximum number of Qizmt_Log calls for each job type by: qizmt maxuserlogsupdate <integer>

If there are millions of error conditions in the data that you need to list, then don’t use Qizmt_Log but rather can make a separate mapreducer to find these. public virtual void Map(ByteSlice line, MapOutput output) { . . . if(error_condition) { output.add(… some error in data }

Why am I getting "access denied" errors accessing resources from a job even though permission has been granted?

If permissions have been modified on the account that Qizmt was installed under, a Qizmt killall command will need to be issued to update the access token.