🪵 Latch Changelog — August 2024
-
Registry:
- Introduced fine-grained Registry access control. On a per table level, Admins can now specify whether their Members have “read-only” or “read-write” access to Registry tables.
-
Data:
- Clicking on an H5AD file on Latch now opens a new Pods with CELLxGENE! Latch Pods are also dynamically sized based on the input file size.
CELLxGENE - LatchBio
-
Workflows:
- Developers can now associate a GitHub commit hash to every registered Latch Workflow’s version
Versioning - LatchBio
- Add support for p3.8xlarge, p3.16xlarge, and p3.24xlarge as new V100 GPUs resource profiles for workflow tasks, leading to significant speed improvements for Dorado base calling workflows.
-
Nextflow:
- Display the file storage used for every Nextflow execution to increase cost transparency
-
Pods:
- Changed the defaults of Jupyter Lab to hide .sysbox files by default and enable deletes of non-empty directories.
- Each user can now specify a personal password for their individual Pods.
- Patched the source code of RStudio to enable reconnecting to RStudio sessions when the code is running. Please see interactive videos of the “Before” and “After” state of RStudio here New RStudio Improvement
- Expanded the number of AWS regions for Latch Pods, resulting in speed improvements for requesting NVIDIA A10 GPU Pods.
- Introduced significant improvements to speed and stability of LData FUSE:
- Sped up file and directory operations
- Fixed concurrency issues in Python sciprts
-
Plots:
- Dependencies and resource profiles can be saved in Plot Templates.
- Enabled GitHub-flavored spoiler tags in Markdown cells.
- Introduced various UX improvements to make Plots more intuitive for first-time users
🪵 Latch Changelog — July 2024
- Data:
- Added a
--cores
flag to latch cp
to enable developers to utilize the appropriate number of cores on their computer for efficient upload and downloads of 1TB+ of data
- Expanded support of Latch’s native IGV browser for VCF, BAM, and bigwig files.
- Registry:
- For Benchling syncs to Registry, file names are now imported instead of file IDs.
- Introduced an API to allow developers to retrieve Registry tables as dataframes with
Registry(id=<id>).get_dataframe()
- Snakemake:
- Better Flyte error messages parsing and display in Snakemake
- Workflows:
- Developers can now specify a
.latch/config
file in the workflow’s directory to specify the workspace that workflow will be registered to.
- Introduced a new “Results” tab to each workflow execution to help developers better surface key result files to scientists.
- Nextflow:
- Sample IDs are now displayed on the UI for batched proceses.
- Improved the UX of zoomable DAGs of Nextflow processes
- Introduced ability for developers to view logs from past retries
- Added EFS as the shared file storage system across all processes. Exposed to developers the option configure how long EFS should be kept around to facilitate relaunching from failed tasks.
- Plots:
- Python cells in Plots can now programmatically output interactive Plotly figures and static images.
- Latch Plots now use Latch Pods’ infrastructure, which means:
- Developers can SSH into the Plots’ underlying computer
- Plots’ computers can be scaled up and down easily up to 93 CPU cores, 1918 GiB RAM, 16,000 GB of storage space, and 8 NVIDIA A10G GPUS with 24 GiB VRAM each.
- Improve stability of Plots with:
- Better status indicators on the state of the plot’s runtime kernel
- Disabling widgets when Plots is connecting
- Multiple bug fixes to ensure smooth connection to Plots
- Sped up Plots’ runtime creation (It should now take <30 seconds for a new Plot notebook to find a fresh computer with the requested resources)
- Platform:
- Admins can set credits threshold for each workspace to receive email alerts when the credits cap is hit.
🪵 Latch Changelog — June 2024