Alexander Clausen

Efficient 'stream'-processing

6

Hi, I'm currently evaluating TileDB for [our project](https://github.com/liberTEM/LiberTEM/) and I'm trying to find out how to efficiently read data. So let's say we create a `DenseArray` like this: ```python d_xy...

Running in gdb, I got the following stack trace: ``` (gdb) run -c 'import mkl_fft' The program being debugged has been started already. Start it from the beginning? (y or...

Release LiberTEM v0.10

13

# Release checklist Issues and pull requests to be considered for this release: * https://github.com/LiberTEM/LiberTEM/pull/1286 * https://github.com/LiberTEM/LiberTEM/pull/1267 * https://github.com/LiberTEM/LiberTEM/pull/1283 * https://github.com/LiberTEM/LiberTEM/pull/1289 * https://github.com/LiberTEM/LiberTEM/pull/1291 * https://github.com/LiberTEM/LiberTEM/pull/1293 * https://github.com/LiberTEM/LiberTEM/pull/1298 https://github.com/LiberTEM/LiberTEM/issues/1271 is planned...

Proper FFT backend selection

4

Need to have a more central point of selecting the FFT backend, instead of sprinkling the code with `try..import..except` statements. See also: Comment in blobfinder: https://github.com/LiberTEM/LiberTEM/blob/7da5db3048b9bfeb2720c2e113348d5f4ca8a34a/src/libertem/udf/blobfinder.py#L14 HPC FFT lib fluidfft:...

GUI/client dev on Firefox stuck in reconnect loop

2

See #870 for discussion of the problem. WebSocket doesn't connect, stuck in a reconnect loop. Root cause is still unknown, needs to be debugged.

Controlled pre-fetching

3

Using `madvise`, one can tell the kernel that the memory in a region will be needed, using `MADV_WILLNEED`. Or for file descriptors, the same can be done with `fadvise64` and...

Improve flow control for websocket connection

4

The flow control was already improved recently (#369, #367), but it is not completely done yet. One reason we still send too much data in case of a bad connection:...

`libertem-server`: handle subsequent Ctrl+C presses with more force

1

Currently, it's possible that the `libertem-server` gets stuck, for whatever reason (example: if the dask scheduler listens on an IP that the computer no longer has). It would be nice...

Allow integer arguments for `cpus` and `cudas` in `make_spec`

1

See discussion in #1267 - small usability fix for both dask and pipelined executor.

Error output: ``` 2022-07-23T09:21:26.2489276Z ==================================== ERRORS ==================================== 2022-07-23T09:21:26.2489940Z _________________ ERROR collecting benchmarks/io/test_dask.py __________________ 2022-07-23T09:21:26.2490590Z benchmarks/io/test_dask.py:34: in 2022-07-23T09:21:26.2490918Z class TestChunking: 2022-07-23T09:21:26.2491232Z benchmarks/io/test_dask.py:40: in TestChunking 2022-07-23T09:21:26.2491553Z __nblock_params, 2022-07-23T09:21:26.2492600Z E NameError: name '_TestChunking__nblock_params'...