fix(atelet): stream sandbox asset downloads instead of buffering in memory#281
Open
Davanum Srinivas (dims) wants to merge 1 commit into
Open
Conversation
…emory Previously, fetchAsset retrieved each asset through FetchFromGCS, which reads the entire object into a byte slice before hashing and writing it. Peak memory therefore scaled with the asset size. This was acceptable for the roughly 50 MB runsc binary, but the micro-VM runtime's kernel and root filesystem images range from hundreds of megabytes to several gigabytes and would exhaust the memory of atelet, which is shared across all actors on a node. The download now streams directly to the temporary file, computing the SHA-256 digest in the same pass via io.MultiWriter and bounding the transfer with an io.LimitReader. Peak memory is now the size of the io.Copy buffer, independent of the asset size. The LimitReader also provides a secondary safeguard for disk usage against a misconfigured or malicious URL that returns an unbounded stream. The limit is 8 GiB and is declared as a variable so that tests may lower it. The digest is verified only after the copy completes. A size or hash failure therefore leaves the data at the temporary path, which is subsequently removed; it is never renamed to the content-addressed cache path, so a failed download cannot corrupt the cache. This change also introduces ategcs.Open, a streaming reader, along with the accompanying TestFetchAssetStreaming.
Collaborator
Author
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Previously, fetchAsset retrieved each asset through FetchFromGCS, which reads the entire object into a byte slice before hashing and writing it. Peak memory therefore scaled with the asset size. This was acceptable for the roughly 50 MB runsc binary, but when we go to micro-VM(s) the runtime's kernel and root filesystem images range from hundreds of megabytes to several gigabytes and would exhaust the memory of atelet, which is shared across all actors on a node.
The download now streams directly to the temporary file, computing the SHA-256 digest in the same pass via io.MultiWriter and bounding the transfer with an io.LimitReader. Peak memory is now the size of the io.Copy buffer, independent of the asset size. The LimitReader also provides a secondary safeguard for disk usage against a misconfigured or malicious URL that returns an unbounded stream. The limit is 8 GiB and is declared as a variable so that tests may lower it.
The digest is verified only after the copy completes. A size or hash failure therefore leaves the data at the temporary path, which is subsequently removed; it is never renamed to the content-addressed cache path, so a failed download cannot corrupt the cache.
This change also introduces ategcs.Open, a streaming reader, along with the accompanying TestFetchAssetStreaming.
Fixes #<issue_number_goes_here>