Bill Katz
Bill Katz
Keep persistent write-ahead log for long-lived ops like split, merge, and data syncs via pub/sub that would allow cleanup after restart of crash. Some ops would be kept, like split/merge,...
Some tasks can be done when DVID is not being actively used by interactive clients. This is currently handled by throttling and the "interactive" query flag. A better system should...
There was a rare dvid crash using gbucket. Here's the little we had on the crash dump: ``` panic: runtime error: invalid memory address or nil pointer dereference [signal SIGSEGV:...
Ting noted that after embedding /raw endpoint data from googlevoxels into local grayscale data, there appears to be misalignment. http://emdata2.int.janelia.org:7000/api/node/e2f0/google_grayscale/raw/0_1_2/513_513_1/5913_6795_5122
At high concurrency and large data access, single file system leveldb backends can be thrashed. For the storage/leveldb drivers, it would make sense to serialize access requests via an I/O...
Now that we are trying different internal block sizes, review code to make sure interplay between data instances with different underlying block sizes work together properly. In particular: - ROI...
Create datatype that keeps track of top N labels in terms of voxels and number of synaptic elements. Presumptive datatype name = labelindex.
During asynchronous denormalizations, if an error might cause inconsistency among denormalizations, have another class of error messages, e.g., dvid.SyncErrorf() instead of dvid.Errorf(), so we can easily detect if any such...
push with transmit=flatten works as intended, but the default transmit all pushes no kv pairs. Add test framework for push between two dvid instances with an eye toward benchmarking.
If a data instance implements a spatially-delimited filter (e.g., ROI), dvid will iterate through all key-value pairs, check if it can be skipped, then pushes the data if it cannot...