LMDB Crash-consistency

[Howard Chu <hyc@symas.com>]

Just an fyi... As you would know from reading the LMDB design papers
( http://symas.com/mdb/#pubs ) LMDB is crash-proof by design. A Symas client 
already confirmed this in their own crash testing last year 
https://symas.com/carrier-grade-stability-and-performance/ and it has again 
been verified by a research group at the University of Wisconsin. Their 
findings are being presented at the Usenix OSDI conference this week, and you 
can read the paper here 
https://www.usenix.org/conference/osdi14/technical-sessions/presentation/pillai

They report on a single "vulnerability" in LMDB, in which LMDB depends on the 
atomicity of a single sector 106-byte write for its transaction commit 
semantics. Their claim is that not all storage devices may guarantee the 
atomicity of such a write. While I myself filed an ITS on this very topic a 
year ago, http://www.openldap.org/its/index.cgi/Incoming?id=7668

the reality is that all storage devices made in the past 20+ years actually do 
guarantee atomicity of single-sector writes. You would have to rewind back to 
30 years at least, to find a HDD where this is not true.

The UWisc researchers' point is that we cannot say what behaviors will be 
exported by up--and-coming nonvolatile RAM mechanisms (e.g. MRAM or PCRAM); if 
they offer byte-addressability instead of sector-addressability then there's a 
potential for these writes to become non-atomic in the future.

At any rate, this issue has zero relevance today, and we are monitoring all of 
the upcoming NVRAM technologies closely for future developments.

The other takeaway from these reports is how critically unreliable many other 
popular systems are. If you use any of the other projects that were included 
in this research, you owe it to yourself to rethink that usage, or raise 
discussions with their developers on how they plan to address their many 
weaknesses.

As an interesting footnote, BerkeleyDB was included in the original testing, 
and while it was in their preliminary results 
http://wisdom.cs.wisc.edu/workshops/spring-14/talks/Thanu.pdf it is now 
conspicuously absent from the final paper. Regardless, the OpenLDAP Project is 
deprecating use of BerkeleyDB for multiple reasons. Again, if you're still 
using BDB you need to take a moment to re-evaluate your project.

--
  -- Howard Chu
  CTO, Symas Corp.           http://www.symas.com
  Director, Highland Sun     http://highlandsun.com/hyc/
  Chief Architect, OpenLDAP  http://www.openldap.org/project/