There's a number of major features all coming together over the next
weeks, so I wanted to raise some awareness of possible areas of
interaction. My concern is more than just "will multiple patches apply
together?", but extends into whether the features interact with each
other with synergy, ignore each other or have negative impact.
The projects that concern me are the following internals projects
Plan Invalidation
Autovacuum changes
Bitmap indexes
GIT indexes
Dead Space Map (DSM)
HOT Updating
Updateable cursors
Restartable VACUUM
The potential interactions as I understand them are these: I'm not
saying this list is either complete or accurate; checking whether it is
or not is exactly the reason for this post.
Anyway, just trying to raise awareness to look for potential issues.
UpdCursors
----------
Syntax changes only, no impact on other projects mentioned.
HOT
---
GIT
Used to be some conflicts at patch level, now gone?
Positive impact on GIT performance
PlanInval
heap_check_hotupdate() test can performed in planner, so
will no longer be a heap function
DSM
HOT updates need not be logged, since they will be cleared
by retail VACUUMs
BitmapIdx
Potential issue with chilling tuples during CREATE INDEX
AutoVac
HOT will change how UPDATEs are reported to pg_stat, by
adding additional fields.
UpdCursors
Should tidscan follow chain to correct tuple? That seems to
change current behaviour even if it might be useful.
Plan Invalidation
-----------------
No known changes, other than the integrations already mentioned
Autovacuum changes
------------------
DSM
Is implementing new forms of VACUUM: VACUUM and VACUUM ALL.
Does reporting of row stats change?
Bitmap indexes
--------------
GIT
Do we need both? I think yes, but need to see some info on
when crossover occurs between two index methods.
Restartable VACUUM
------------------
DSM
Interactions with DSM?
HOT
Must use full cycles only
--
Simon Riggs
EnterpriseDB http://www.enterprisedb.com
---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster
