Skip to content

Commit

Permalink
Doc feedback
Browse files Browse the repository at this point in the history
  • Loading branch information
jgainerdewar committed Feb 11, 2025
1 parent 57ed4b4 commit 1f036c1
Showing 1 changed file with 3 additions and 5 deletions.
8 changes: 3 additions & 5 deletions docs/Configuring.md
Original file line number Diff line number Diff line change
Expand Up @@ -509,8 +509,6 @@ Cromwell will try each algorithm in the order listed until it finds an available
types that aren't guaranteed to have a single type of hash. Cromwell will never download a file in order to compute its
hash, all cloud hashing strategies are dependent on file metadata.

Users may provide an empty list `hashing-strategy: []` to disable call caching for a filesystem.

Each filesystem supports a limited set of hash strategies.
* `gcs` Google Storage supports:
* `crc32c` default, guaranteed to exist
Expand All @@ -534,9 +532,9 @@ is `["crc32c", "md5", "sha256", "etag"]`
For some high-throughput production use cases that run many, many copies of the same task differing by only one input file,
the collision rate of `crc32c` may be unacceptably high. To dramatically reduce the chance of collision at the cost of
reducing the collection of tasks that can be call cached, we recommend `hashing-strategy: ["md5", "identity"]`. This
will use `md5` hashes when they exist, and fall back to the very strict `identity` strategy when they do not. Because
all GCS files created by Cromwell are guaranteed to have `md5`, `identity` comes into play only for user-provided workflow
input files.
will use `md5` hashes when they exist, and fall back to the very strict `identity` strategy when they do not (ex. if
a file was created by multipart upload). Because all GCS files created by Cromwell are guaranteed to have `md5`,
`identity` comes into play only for user-provided workflow input files.

##### Call cache strategy options for local filesystem

Expand Down

0 comments on commit 1f036c1

Please sign in to comment.