Lioran S3 V1 documents a PUT sequence that promotes an object’s payload file before writing its metadata to RocksDB. If that metadata write reports an error, the application attempts to delete the promoted file. But if the process stops after the rename and before the metadata commit, that cleanup cannot run: the result can be orphaned bytes on disk, not a transactionally committed object.
That distinction matters. The project describes specific cleanup paths, not a single atomic transaction across the filesystem and database, and its published account is not an independent code audit or verified crash-test report.
As an Amazon Associate I earn from qualifying purchases.
What the documented PUT pipeline does
Lioran S3 V1 is described by its project as a self-hosted, single-node object-storage engine. Payloads live on the filesystem; object metadata is stored in RocksDB. The project calls the release Pre-Alpha and says it uses its own REST interface rather than serving as a drop-in AWS S3-compatible layer or a distributed object store. Project overview
In the project-authored walkthrough, a PUT proceeds through these stages:
#1 Best Overall
- Validate the bucket and object key.
- Check capacity and quota.
- Create a staging file.
- Stream the incoming payload into staging while calculating its SHA-256 hash.
- Flush the staged data; in strict mode, perform an explicit synchronization step.
- Close the staging file, then check capacity and quota again using the payload’s actual size.
- Create the destination’s parent directory.
- Rename the staging file to the final object path.
- Construct the object metadata and write it to RocksDB.
- Return success, or, if the metadata write reports an error, attempt to remove the promoted payload.
The consequential boundary is between steps eight and nine: the payload reaches its final filesystem path before its metadata record is written. Project-authored PUT walkthrough
What happens at each failure stage?
| Failure stage | Documented response | What may remain |
|---|---|---|
| Incoming stream fails | The project says it removes the staging file. | No committed metadata record has been written; the described metadata-driven read path does not treat the failed upload as a committed object. |
| Second capacity or quota check rejects the upload | The project says it removes staging after learning the exact payload size. | The payload has not been promoted and metadata has not been committed. |
| Rename fails | The call returns a storage error and attempts staging-file cleanup. | Metadata has not yet been committed. |
| Metadata write reports an error | The payload is already promoted; the application attempts to delete it as rollback. | Cleanup is an attempted response to a synchronous error, not a guarantee that deletion succeeds. |
| Process stops after rename but before metadata commit | The process cannot execute the rollback code after it has stopped. | Payload bytes can remain at the final path without a committed metadata record: an orphan that may consume disk. |
These responses are the project’s description of the implementation, not independently verified outcomes for every failure mode. Project-authored PUT walkthrough
Rank #2
If the process crashes between rename and metadata commit
The rename has promoted the payload, but the RocksDB metadata record has not yet been committed. Because the process is no longer running, its synchronous error-handling path cannot remove the file. The project therefore identifies a possible state in which bytes exist without a corresponding committed object record. Project-authored PUT walkthrough
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesAccording to the project’s account, ordinary reads start from metadata, so this orphaned file is not automatically exposed as a committed object. It can still occupy storage and become a disk-recovery concern. That is a description of the metadata-driven read path, not evidence of an automatic orphan scan or recovery process.
Rank #3
What strict and balanced durability do—and do not—mean
The project’s materials distinguish durability modes by their documented persistence steps: strict mode includes an explicit flush or synchronization boundary before metadata is treated as durable, while the PUT walkthrough describes fsync as optional. The available descriptions do not establish that synchronization is always enabled, that every file and directory metadata change is synchronized, or that all filesystems and storage hardware honor persistence requests identically. PUT walkthrough · Project durability article
Neither mode, as described, makes the filesystem rename and RocksDB metadata write a single atomic transaction. The project’s founder and CTO, Swaraj Puppalwar, summarizes the boundary: “It does not magically turn filesystem + RocksDB into one atomic transaction.” Project-authored failure-window account
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the published account does not establish
The project describes the sequence and its intended cleanup behavior, but the available account does not establish platform-by-platform rename semantics, every concurrent-write or database-failure case, or results from a completed fault-injection campaign. It supplies no attributable reliability rate, benchmark, or independently verified crash-test result for this pipeline. Mentioned stage-level timing instrumentation describes what can be measured; it is not a published latency or throughput result. Project materials on testing and evidence
Accordingly, the useful conclusion is limited but concrete: normal application error handling attempts to clean up in several failure paths, while an abrupt stop in the rename-to-metadata interval can leave orphaned payload bytes. The documented sequence does not support a broader claim that every crash is recovered or that the filesystem and database commit together.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
