WAL questions
What is there for a user or admin or programmer to know about the new WAL
stuff? What all does it do? Does it allow for an audit file to be created,
which can be used to playback and/or rewind the transactions on the database
by user/admin commands? How do checkpoints limit or affect how far back
recovery is possible (if at all)? When is a checkpoint made? Does it allow
for online recovery or only offline? What are the settings/parameters that
control it (if any, like size of log at which to cut off a new one)? Can the
WAL files that are made be read by humans and where are they stored? How
transparent is this feature? Maybe I'm confused!
I have some familiarity with a mainframe database that made audit tapes of
all transactions. A nightly full dump was also made. Kinda old stuff. The
tapes were kept for something like 2 weeks before being recycled. If the
database crashed, it was possible to restore the database back to any time by
using a full dump and some audit tapes. The dumps and audit tapes were
specific to a database, not the whole DBMS. Other databases could be up and
running normally while one was being rolled back and then forward again after
fixing some problem. Even the one being restored could be online, queuing or
processing some queries until the recovery was done. Something like that, it
was a hospital environment. How does WAL compare to any of this, if at all?
Can WAL be described as a deferred fsync of a batch of transactions?
--
-------- Robert B. Easter reaster@comptechnews.com ---------
-- CompTechNews Message Board http://www.comptechnews.com/ --
-- CompTechServ Tech Services http://www.comptechserv.com/ --
---------- http://www.comptechnews.com/~reaster/ ------------
"Robert B. Easter" <reaster@comptechnews.com> writes:
What is there for a user or admin or programmer to know about the new WAL
stuff?
Vadim is the man who ought to answer this (and he's on the hook to write
a lot of documentation before 7.1 ships ;-)). But my understanding is
that as of 7.1, WAL will not really provide any user-level features like
audit trails or point-in-time recovery. The only useful thing it does
right now is reduce the cost of fsyncs. It provides an infrastructure
on which we can build audit trails etc in future releases --- but the
superstructure atop this infrastructure ain't there yet.
Over to you, Vadim ...
regards, tom lane
Vadim is the man who ought to answer this (and he's on the
hook to write a lot of documentation before 7.1 ships ;-)).
But my understanding is that as of 7.1, WAL will not really
provide any user-level features like audit trails or
point-in-time recovery. The only useful thing it does right
now is reduce the cost of fsyncs.
and protects against
- non-atomic disk writes (eg partially written page cleaned
up by vacuum);
- losing tuples in btree split (first step on the way to
stable indices)
It provides an infrastructure on which we can build audit
trails etc in future releases --- but the superstructure
atop this infrastructure ain't there yet.
Exactly.
BTW, WAL related questions should be posted to -hackers list.
-general is not for discussion of upcoming releases.
Vadim
Import Notes
Resolved by subject fallback