Binary data import error with SciDB 14.12


#1

Hi,

for SciDB 14.8, I wrote myself a loading script that converted my NetCDF data to binary, split it into n chunks (where n == number of nodes in the cluster), and load it in parallel with
load(FLATARRAYNAME,LOADFILENAME,-1,‘format’)

I just upgraded to SciDB 14.12 (Ubuntu binary package) and tried throwing away the old DB, re-initialising from scratch and then loading the data with my trusty script. Sadly, the import that worked just fine for 14.8 now gives me this error:

2015-01-21 18:54:27,258 [0x7fd1158ae700] [INFO ]: Loading of ascat_soil_moisture_flat_all is completed: loaded 0 chunks and 0 cells with 0 errors
2015-01-21 18:54:27,258 [0x7fd1158ae700] [ERROR]: ServerMessageHandleJob::handleExecutePhysicalPlan: QueryID = 1100975610688 encountered the error: UserException in file: src/query/ops/input/InputArray.cpp function: moveNext line: 371
Error id: scidb::SCIDB_SE_IMPORT_ERROR::SCIDB_LE_FILE_IMPORT_FAILED
Error description: Import error. Import from file 'ascatbatch' (instance 1) to array 'ascat_soil_moisture_flat_all' failed at line 0, column 0, offset 0, value='': Stepped out of bounds computing next chunk location.
2015-01-21 18:54:27,259 [0x7fd1158ae700] [ERROR]: ServerMessageHandleJob::run: Error occurred in message handler: UserException in file: src/query/ops/input/InputArray.cpp function: moveNext line: 371
Error id: scidb::SCIDB_SE_IMPORT_ERROR::SCIDB_LE_FILE_IMPORT_FAILED
Error description: Import error. Import from file 'ascatbatch' (instance 1) to array 'ascat_soil_moisture_flat_all' failed at line 0, column 0, offset 0, value='': Stepped out of bounds computing next chunk location., messageType = 2, sourceInstance = 0, queryID=1100975610688
2015-01-21 18:54:27,259 [0x7fd1158ae700] [DEBUG]: ServerMessageHandleJob::run: Execution of query 1100975610688 is aborted on worker
2015-01-21 18:54:27,259 [0x7fd1158ae700] [DEBUG]: Query::done: queryID=1100975610688, _commitState=0, errorCode=338

Any hints on what to do about this? I’m sure SciDB is trying to tell me something but I don’t understand the problem What does “Stepped out of bounds computing next chunk location” mean?


#2

Hi, sorry for the difficulty. Can you provide the schema of the target array, and the binary format string used for the save and load? I am trying to repro the issue in-house.

Mike


#3

Sure :smile:
The flat array that I’m trying to load into is created like this:

create array ascat_soil_moisture_flat_all
<
gpi: int64,
lat: float,
lon: float,
time_scaled: int64,
soilmoisture: uint8,
soilmoisture_noise: uint8,
surface_state_flag: uint8,
processing_flag: uint8,
orbit_dir_asc: bool
>
[i];

The load command looks like this:

load(ascat_soil_moisture_flat_all,'"ascatbatch"',-1,'(int64,float,float,int64,uint8,uint8,uint8,uint8,bool)');

Thanks for looking into this :smile:


#4

Good news, I was able to reproduce the problem, and I have a bug fix. Unfortunately this is a pretty serious bug in parallel loading of all formats other than opaque and SciDB text format. I will have to discuss with management how best to get the fix out to the field.

In the short term: I will run some regression tests overnight (and maybe add a new one :frowning: ). Then if you build from source, I can get you a patch. If not, let me know what OS you are running and I will build you an unofficial binary.

Mike


#5

That’s nice to hear :smile:

Does this mean it may be possible to export the data from a SciDB 14.8 to opaque format and then load this into a new SciDB 14.12 as a workaround?

I’m working with an Ubuntu 12.04 machine and installed the binaries from the APT repository (deb downloads.paradigm4.com/ ubuntu12.04/14.12/). Your help is much appreciated :smile:


#6

Yes, opaque format should work. Out of paranoia, I just did a small test here to verify it. (Our aim is always to allow opaque save/restores across successive versions… but no promises if you try to skip a version).

Let me know how that goes.

Mike


#7

Breaking news: we are now planning to respin the release to address this issue. Stay tuned for announcements. The new bits should be ready this week, depending on how the coming snow storm affects our power.


#8

Wow, it seems I triggered a fair bit of work for you - sorry about that and thanks for not shooting the messenger. :smile:

On the opaque front it’s looking good - I tried importing from a SciDB 14.8 backup and it works so far, so I can now enjoy my new SciDB 14.12.

Good luck with the weather and the respin…