| Age | Commit message (Collapse) | Author |
|---|
|
If the remote server cannot be connected and query is canceled,
don't show random warning about "Unfinished connection".
Patch by Fazal Majid
|
|
Without it, the test can fail occasionally.
Reported by Peter Eisentraut.
|
|
Otherwise it can still hang.
|
|
the validator is too big change for 2.5.1
|
|
|
|
Previous check was too broad and left plproxy hanging.
Reported-By: Tarvi Pillessaar
|
|
Refactor some internals to make this possible. Mainly, FunctionCallInfo
is not available when validating, so avoid accessing that if not
required. Rename plproxy_compile() to plproxy_compile_and_cache() and
the previously internal fn_compile() to plproxy_compile(). This matches
their purpose better and allows the validator to call plproxy_compile()
without invoking execution-time dependent code.
Many error test cases have changed because the validator catches errors
when the function is created, not when it is called.
Raise the extension version to 2.5.1 to be able to upgrade from
non-validator installations.
|
|
|
|
|
|
* use installed pg_buildext
* use installed pgxs_debian_control.mk
* don't override 'clean' target in debian/rules
* 'make deb' forces control file regeneration
This reduces crap maintained locally.
It also does mean that to build against server-dev-X.Y package
that is not for Debians default Postgres version, following packages
need to be installed from PGDG (wiki.postgresql.org/wiki/Apt) repo:
postgresql-client-common
postgresql-common
postgresql-server-dev-all
|
|
|
|
Fix build with PostgreSQL 9.3devel
|
|
Some needed declarations have been moved to the new header
htup_details.h.
stringinfo.h needs to be included explicitly now. It should always have
been necessary, but might have come in through other headers.
|
|
|
|
|
|
|
|
|
|
Also throw error on unknown proargmode.
Fixes crash reported by Sébastien Lardière.
|
|
|
|
|
|
|
|
|
|
Previously, as soon as cancel requests were send,
plproxy re-throwed the error, without waiting for
reaction from backend. Such behaviour creates
2 problems:
- If plproxy backend is closed immediately, the bouncer will
see plproxy close before cancel from backend, thus seeing
mid-tx close, thus dropping the connection.
- If new query comes in to plproxy backend, plproxy itself
will see dirty connection, closing it, thus also causing
close of server connection in bouncer.
In both cases it can cause server connection drop in pooler.
New behaviour of waiting query result should fix it.
|
|
On each call, recheck type relation rowstamp.
If it proves to be noticeable, we can use
syscache callback infrastructure for that,
but it will be much more complex.
|
|
|
|
|
|
This avoids breakage if pg_config is not found
and PGVER stays empty.
|
|
|
|
Otherwise there can be garbage in struct
|
|
|
|
Direct casts hardwire the assumption that node
is always first member of struct.
|
|
|
|
|
|
|
|
|
|
This allows to fall back to old default of 'session_user'.
New default is 'current_user'.
|
|
This allows falling back to defaults when
some value is dropped from config.
|
|
|
|
Now connections are added to active_list in order
they get tagged from hash function.
|
|
Plus minor cleanups.
|
|
|
|
|
|
|
|
|
|
|
|
- Add aatree.[ch]
- Dont run maintenance if no init is done
- Simplify header deps in Makefile
- Store fake cluster in binary tree
|
|
|
|
- remove plproxy.so even when building something else
- Use ctags for tags
|
|
|
|
Drop yada based packaging, instead use the new pg_buildext
packaging, except include pg_buildext under debian/ so the
package builds on lenny too.
It finds any postgresql-server-dev packages and builds
against them.
Also import Peter Eisentraut's Makefile fixes from official
Debian packaging to support multi-package builds.
Now package is built for all installed server-dev versions.
|
