[Info-vax] PostgreSQL, POSIX, SSIO, OpenVMS data corruption (was: Re: [From VSI website] VSI Announces a Strategic Partnership with Sector 7)
Stephen Hoffman
seaohveh at hoffmanlabs.invalid
Thu Jan 28 13:23:13 EST 2021
On 2021-01-28 15:44:11 +0000, Richard Murphy said:
> There is not a version of Postgres database server for OpenVMS and I
> believe there cannot ever be due to some Postgres internals
> complexities.
If by "Postgres internals complexities", you mean that the PostgreSQL
port is wedged behind an OpenVMS user data corruption, yes.
This corruption was known as the Shared Stream I/O SSIO bug, having
been researched by HP OpenVMS development.
The following HP presentation has an overview of the SSIO corruption:
http://de.openvms.org/TUD2012/opensource_and_unix_portability.pdf
Highlights from HP:
> On OpenVMS, concurrent POSIX write() calls to the same file can corrupt data
> - POSIX I/O on OpenVMS not atomic
> − Data updates can get lost
> − Disk can get mixed data from overlapping writes
> Consistency not guaranteed for files opened for shared write
> Victims: UNIX programs ported to OpenVMS
> - Programs must provide atomicity on their own
> Stated formally:
> — OpenVMS does not provide POSIX-compliant shared read/write to byte
> stream files
Contrary to the above slide, it'll hit other users of the write() call,
not just ported apps.
Apropos of little else here and only slightly related to this posting:
https://posixtest.sourceforge.net
--
Pure Personal Opinion | HoffmanLabs LLC
More information about the Info-vax
mailing list