Known Limitations

By design, Quilt is backed by Amazon S3 and scales to billions of objects and petabytes of data. The underlying limitations of S3 apply.

The Quilt catalog can browse packages and S3 buckets of any size.

Catalog push

To ensure usability and quick package pushes, the Quilt web catalog imposes the following limits on pushes (which vary depending on whether the chunked checksums are enabled on the stack). These limits do not apply to the quilt3 Python API.

Dimension	Max (classic / chunked checksums)
Package manifest size (metadata)	100 MiB
Package size (data; via promotion or from an S3 directory)	100 GiB / 5 TiB
Total size of uploaded files (soft limit)	20 GB
Total size of files from S3 (soft limit)	50 GB / 5 TB
Maximum file size	10 GiB / 5 TiB
Maximum number of files per push (soft limit)	1,000
Maximum number of files per push (hard limit)	5,000

Important note about S3 prefix enumeration

When creating a package from an S3 prefix (folder) using the Catalog GUI, the interface will display a warning if the prefix contains more than 1,000 files (the soft limit): "Total number of files exceeds recommended maximum of 1000." You can proceed past this warning, but the number of files the Catalog enumerates from S3 may vary, and browser performance may degrade with large file counts. The hard limit is 5,000 files per push (see table above). Additionally, S3 does not guarantee a stable ordering of listed objects, so the exact set of files included may differ between attempts.

Workarounds for large directories:

• Use the quilt3 Python API's Package.set_dir() method, which has no such limitation

• Use the Packaging Engine to automate package creation from S3 prefixes

• Split your data into multiple packages, each under the 1,000 file limit

API

As of this writing, with sufficient client-side memory, you can comfortable scale Quilt packages to at least one million objects per package, with no practical limit on object size (save S3's 5 TB per object limit). A fast network, or better yet an AWS compute instance in the same region as your Quilt S3 buckets, is recommended.

Metadata

See Metadata for teams.