Repository navigation
[doc] Clarify meaning of "crash" in transaction consistency guarantee #281
Description
Hi, thanks for your great work on libmdbx!
I have a question regarding the documentation, specifically this part:
"1,000,000 transactions in async-write mode. In case of a crash all data is consistent and conforms to the one of last successful transactions, but lost transaction count is much higher than in lazy-write mode. All DB engines in this mode do as little writes as possible on persistent storage. libmdbx uses msync(MS_ASYNC) in this mode."
Could you clarify what exactly is meant by "crash" in this context? Do you mean:
- A process crash (e.g., the database process is terminated unexpectedly)?
- A system crash (e.g., OS kernel panic or power loss)?
I ask because libmdbx in this mode uses msync(MS_ASYNC), which (as I understand it) ensures data is passed to the kernel page cache but does not guarantee it is flushed to persistent storage immediately. So in the case of a process crash, the data should still be safe, but in the case of a system crash or power failure, data loss could occur unless a stronger sync is used.
Could you confirm whether the "crash" here includes system-level crashes, or if it's limited to process-level failures?
Thanks a lot in advance!
Activity
erthink commentedon May 10, 2025
I agree, the paragraph should be rewritten (it looks like some kind of not-a-good machine translation option has been preserved).
For now, please see the SYNC MODES and subordinary references, especially for
MDBX_SAFE_NOSYNCandMDBX_UTTERLY_NOSYNC.erthink commentedon May 12, 2025
After thinking about it, I decided to give an explanation, as apparently it's time to remove some information from the README due to obsolescence.
Most (or all) of slides that now within REAME was created/prepared about of 10 year ago for Highload++ conference. At that time, libmdbx differed relatively little from LMDB. The current version of libmdbx offers/supports a different set of sync-to-disk modes, so the slide descriptions are somewhat outdated.
The noted "crash" mean a system crash (e.g. OS kernel panic or power loss), and your understanding is correct.
mdbx-doc: clarify "crash" meaning.
erthink commentedon Aug 15, 2026
This topic has been included in the FAQ, and AI has already collected all the facts and explanations related to it.
mdbx-doc: clarify "crash" meaning (backport).
mdbx-doc: clarify "crash" meaning (backport).