Direct multi-node archiving: Each node can upload log segments, indexes, and other files directly to the archive storage classes of AWS S3, Alibaba Cloud OSS, Volcengine TOS, or Google Cloud Storage. The online catalog, publish receipts, and account authorization are stored in a separate standard storage prefix within the same bucket, which eliminates data copies in shared hot storage.
Direct archiving supported in the initialization wizard: The wizard checks Redis and verifies whether the two storage prefixes overlap; before saving, it displays the selected storage class and whether queries require thawing. Directly readable storage classes support cross-node queries, while frozen classes follow the approval and thawing process, and newly written logs also follow this rule after archiving.
Extended shared account state storage: OSS, GCS, and TOS with WORM enabled save accounts and policy groups through atomic creation of immutable versions, preventing concurrent operations from overwriting each other's authorization information.
S3 multipart resumable upload: Validation is performed against the original part size of the uploaded object. After adjusting multipart_part_bytes and restarting, uploaded objects can still be reused.
Archive integrity verification: When a directly readable object lacks a server-side checksum, a limited streaming read-back is used to compute BLAKE3, eliminating the need for an extra local copy.
Reclamation of unpublished objects: The maintenance lease is reconfirmed before each deletion, and the grace period is included in the archive-side rate limiting. If cleanup is interrupted, the upload intent is retained so that subsequent maintenance nodes can continue processing.