We add information to the TwoPhaseHeaderFile about which tablespace we're
attempting to drop. We also schedule a tablespace for deletion (similar
to CREATE TABLESPACE) on commit, which waits until the transaction
is committed to perform the filesystem deletion. This follows the
recommendation for relation files to be deleted after the commit to not
spend time deleting files during the commit.

We also modified the distributed_commit record to pass the tablespace to
be dropped to the standby. And, we modified the abort record, to help
back out of a drop on the mirrors.

- Bump the catalog for the drop tablespace xlog changes.
- Extracts hook functions to be implemented by separate components

The tests skip FTS probes during the test run to avoid FTS taking a primary
down that has a fault point injected into it.
Co-authored-by: NTaylor Vesely <tvesely@pivotal.io>
Co-authored-by: NSoumyadeep Chakraborty <sochakraborty@pivotal.io>
Co-authored-by: NAlexandra Wang <lewang@pivotal.io>

Setting a variable to itself is a no-op which can be removed. This
may have been introduced in error and instead masking a real bug,
but if it so then we have lived with it for two years so I'm opting
for removing.

Reviewed-by: Asim R P and Bhuvnesh Chaudhary

Reviewed-by: David Yozie

This test is run in a parallel group and uses table names that collide
with other tests that either run before it, or concurrent with it. This
leads to difficult-to-debug CI failures. Commit 95a33fb4 already
address a similar issue in two other tests in the same group, but
somehow missed this.

This commit schema qualifies all objects in that test.
Co-authored-by: NTaylor Vesely <tvesely@pivotal.io>

* docs - [pg->gp]_dist_wait_status UDF name change

* add sample output from simon

In fact, if you manage to get an extra directory created, you can
exercise the code being removed. It won't work anyway, with an error:

> make[1]: Entering directory '/build/gpdb/gpcontrib'
> make[1]: *** installcheck: No such file or directory.  Stop.
> make[1]: Leaving directory '/build/gpdb/gpcontrib'
> make: *** [Makefile:96: installcheck] Error 2

This should have been deleted in commit ad6b920f as part of GPHDFS
removal.

background worker is not scheduled until distributed transactions
are recovered if it needs to start at BgWorkerStart_RecoveryFinished
or BgWorkerStart_ConsistentState because it's not safe to do a read
or write if DTX is not recovered. GPDB is designed to do this check
in master only, however Gp_role == GP_ROLE_DISPATCH is not a
sufficient check for master.

Spotted by Wang Hao <haowang@pivotal.io>

Co-authored-by: NSoumyadeep Chakraborty <sochakraborty@pivotal.io>

High-level details:
1) Tear out existing 2PC infrastructure for databases
2) Introduce new structures (in-mem, on-disk) for making databases 2PC
3) Make sure ALTER DATABASE SET TABLESPACE (ADST) works correctly such
that no orphaned files remain in the source/target tablespace
directories in cases of commit/abort.
4) We add the dboid dir to pendingDbDeletes for abort after copying
dboid dir in movedb. This ensures that the dboid dir is cleaned up from
the target tablespace when we have a failure during the course of
execution of the ALTER.
5) Re-engineer tests to assert filesystem changes and accommodate
multiple failure scenarios during the course of a 2PC commit/abort.
6) Introduce a new generic mechanism for making mirrors catch up. This
is done by inserting a XLOG_XACT_NOOP record and then waiting for the
mirror to replay the record with gp_inject_fault2. The premise is that
if the mirror has replayed this latest record, it would have replayed
everything before the NOOP record. We have introduced a UDF and a fault
injection point to make this possible.
7) Update pg_xlogdump to dump out pending DB deletes on commit/abort.

Refactoring Notes:
1) The xlog insert for database drops in `DropDatabaseDirectory`
is no longer required since we are already WALing database drops
elsewhere.
2) `xact_get_distributed_info_from_commit` populated variables that were
useful in the context of its only caller `xact_redo_distributed_commit`.
If we keep it around there would be unnecessary code duplication -> thus
deleting it.
3) Use DropRelationFiles() instead of duplicating logic in
xact_redo_distributed_commit.
4) Refactor session-level lock release for the QD during movedb().
5) smgr is a layer that pertains only to relations. So, we drop the
smgr prefix for pending db delete functions.
6) DropDatabaseDirectories() pertains to dropping dboid dirs at the
filesystem level. This means that it belongs inside storage.h/c.
DropDatabaseDirectories forms the visible interface as contrasted to
DropDatabaseDirectory. So moved the latter and made it static.
7) Rename DatabaseDropStorage to ScheduleDbDirDelete to call out its
intent.
8) Extract pending db deletes and dboid dir removal to dedicated file
9) Tests for scenarios with and without faults now live in one
consolidated file.
Co-authored-by: NTaylor Vesely <tvesely@pivotal.io>
Co-authored-by: NAdam Berlin <aberlin@pivotal.io>
Co-authored-by: NJesse Zhang <jzhang@pivotal.io>

It is possible, and does happen, that CreateExternalStmt will be written upon
during transformations. The transformer functions are not explicit regarding
ownership and some of them might (and do) invoke walkers that can (and do) write
on the transformed statement. However, some callers of these functions, do pass
arguments that should be concidered read only. Such an example is passing a
CreateExternalStmt which belongs to, and taken from, a cached plan which will be
correctly retained for subsequent executions.

Create a short lived memory context in the transformCreateExternalStmt, which
will additionaly protect against leaks in long expanding statments, and operate
in a copy of the passed statement.
Reviewed-by: NAsim R P <apraveen@pivotal.io>
Reviewed-by: NBaiShaoqi <sbai@pivotal.io>

The macro IS_FILL_WORD was generating a false positive array-bounds
warning. Convting the macro into an inline function removes the warning
without altering the essence of the code. The macro was always used within
a conditional so it is unlikely to suffer performance penalties.

* Revert "CI: skip CLI behave tests that currently fail on ubuntu18.04"

This reverts commit 6a0cb2c6.
This commit is to re-enable behave test for centos/ubuntu

* Ouput the same error message format

gppkg for rpm and deb outputs error messages when consistency is broken, but
the message is not the same, which make a pipeline fail.

* Update gppkg for behave test and fix some minor error in gppkg

Now, the binary sample.gppkg is removed, instead we add sample-*.rpm and sample-*.deb.
Because sample.gppkg is platform specific and GP major version sensitive. The uploaded
rpm/deb files are available unless the rpm/deb file is incompatible on the specific
platform. GP_MAJORVERSION is dynamically retrieved from a makefile installed in the gpdb
folder, so the gp major version in the gppkg will be always correct.
sample.gppkg will only be generated when gppkg tag is provided

The kerberos service name of gpdb and psql must match to allow proper
kerberos authentication. For linux platform this is controlled by
--with-krb-srvnam. For Windows platform it's hardcoded in template
pg_config.h.win32. We currently build gpdb server without specifying
--with-krb-srvnam, and the default value is "postgres". Change the
correspond value that hardcoded Windows config template to "postgres"
also.

- Package necessary kerberos utility with windows installer
- Add Kerberos auth test for Windows client

As discussed in #6932, the size of interconnect connection hash table is 16,
which would be too small for large cluster and could bring too many hash table
collisions. Set to (2 * number segments).

Sanity check update/delete to prevent potential data corruption when there are tuples with bad distribution. (#7304)

Data being stored with bad distribution is not unusual in a real production
environment. There could be issues if the ctid for update/delete is distributed
to another segment. This patch will double check this and error out if needed.

Reviewed-by: Jinbao Chen

* extend background worker framework for old auxiliary process

GPDB used to have its own framework to manage its auxiliary processes like FTS,
GDD, DtxRecovery, Perfmon, Backoff, etc.

This old framework a few defects:
1. GPDB only, different concept and make code different from upstream in many places.
2. Not easy enough to add a new type of auxiliary process although it has tried
   to reduce the complex, but still lots of additional code needed out of core feature.
3. not clear for backend initialization and many duplicated code.
4. not clear for the components/resource needed.
5. using SIGUSR2 to shutdown which differs from upstream.
6. special code for auxiliary in the workflow.

Actually, the old framework is very similar to the background worker framework
in upstream. Now we have bgworker feature merged, it's time to covert those GPDB
processes to background worker now and take advantage of the background worker
feature.

To do this, we need to extend the background worker framework a little bit:
1. extend bgw_start_time
 typedef enum
 {
	/* can start if startup process is lunched*/
        BgWorkerStart_PostmasterStart,
	/* can start basicly when it's saft to read (in GPDB, after DTX recovery is done) */
        BgWorkerStart_ConsistentState,
	/* can start basicly when it's safe to read/write (in GPDB, after DTX recovery is done) */
        BgWorkerStart_RecoveryFinished
+++     BgWorkerStart_DtxRecovering,
 } BgWorkerStartTime;

A new type called BgWorkerStart_DtxRecovering which means a worker can be started
after the startup process finished and before DTX recovery is done. This is
important because DTX recovery process is also a bgworker now and it also need
FTS prober once it hit an error.

2. adding a bgw_start_rule
Some auxiliary processes can only be launched in QD or only be lanuched when
some GUCs are set. So users can provide a custom function to specify the rule
to wether launch a bgworker or not.

* move fts to bgworker
* move gdd to background worker
* move dtx recovery
* move perfmon_segmentinfo to bg worker
* move backoff to bgworker
* move perfmon to bgworker
Reviewed-by: NAsim R P <apraveen@pivotal.io>
Reviewed-by: NHubert Zhang <hzhang@pivotal.io>

Issue is that COPY WITH OIDS requires an OID from the QD for tuple
insert. However, in the case of multi-insert we don't yet have the
multiple OIDs required from the QD. Simplest approach is to disable
multi-insert on COPY WITH OIDS. A better solution may be to send a list
of OIDS from the QD which QE would use in multi-insert.

Sometimes gpfdist with ssl can not start
in 30 seconds.

The mailmap file is used by Git to properly coalesce contributions which
have been made by the same contributor with different email addresses or
user names.

Discussion: https://groups.google.com/a/greenplum.org/forum/#!topic/gpdb-dev/zxD4x6fis0E

Greenplum diverges from upstream PostgreSQL in abort transaction workflow.
Greenplum must dispatch ABORT command to QEs.  Consequently, transaction ID in
top transaction state is reset before broadcasting the ABORT.

Exported snapshots are closely tied to the transaction that exported them.
They must be cleared before the transaction that exported them ceases to exist.
In order to abide by this sequence of operations, Greenplum must also clear
exported snapshots before marking the transaction as invalid.  This patch moves
the call to clear exported snapshots at a location that is right for Greenplum.

Fixes Github issue #8020

Reviewed-by: Georgios Kokolatos and Jimmy Yih

FDW or External table's INSERT actions are actually just some defined
callbacks but not real heap table INSERT actions.

Don't mark transactions doing write just because they insert into
foreign or external tables, some FDW extensions report an error because
they don't see XACT_EVENT_PRE_PREPARE coming.

```
SET "request.header.user-agent" = 'curl/7.29.0';
```

The double quote characters are needed in the above command, but
Greenplum lost them while dispatching, which reports syntax error.

We need to bump the catalog because of the change to the XLOG
introduced by the create tablespace pr.

(122c79f2)

The checkinc.py script was performing a check for the presence of
header files we're not shipping.  This was essentially guaranteed
to always pass as the code was able to compile, and has generally
provided little value.  Recent changes in macOS installations has
also made it fail with false positives, so finally remove it to
minimize developer time spent fighting quite useless tests.

Any checks for headers should be performed in autoconf before the
code is compiled, not during testing.

Discussion: https://github.com/greenplum-db/gpdb/issues/7961
https://groups.google.com/a/greenplum.org/d/msg/gpdb-dev/cCXsHwPr7GM/XbExPStmAgAJ
Reviewed-by: Ashwin Agrawal, Jimmy Yih

To match how bin_gpdb tar.gz is created.
Co-authored-by: NMark Sliva <msliva@pivotal.io>
Co-authored-by: NAmil Khanzada <akhanzada@pivotal.io>

For commit e995ae86Co-authored-by: NMark Sliva <msliva@pivotal.io>
Co-authored-by: NAmil Khanzada <akhanzada@pivotal.io>

ICG changes for Create LAS Correlated Apply on in filter context

Prior to this change it would only run on centos7.
Also pipelines that did not target centos7 would
fail to run.
Co-authored-by: NMark Sliva <msliva@pivotal.io>
Co-authored-by: NAmil Khanzada <akhanzada@pivotal.io>

This reverts commit c46bf52f.

* Adding reference topic and additional information around temp_tablespaces

* Clarify comma-separated list

Fix gen_pipeline.py OS issue

Modify TwoPhaseFileHeader to include a tablespace OID that needs to be
removed if an abort occurs.

We add a 'pending tablespace for deletion' that is scheduled during the
'create tablespace' command. If a node encounters an error, it can abort
using the in-memory value. The master then can send an ABORT PREPARED for all
segments that have prepared the create tablespace.

A successful 'create tablespace' needs to unschedule the deletion, or following
aborted transactions will accidentally delete the newly created tablespace.

XLOG records are written for the ABORT and ABORT_PREPARED of a transaction, and these records also
now contain a tablespace oid for deletion.

Notes:

During CREATE TABLESPACE, if an error occurs after the XLOG_XACT_PREPARE
record has been written, there should be no filesystem changes left
on disk for any of the segments. (master, standby, primary, or mirror)

If an error occurs after the XLOG_XACT_DISTRIBUTED_COMMIT, then the create
tablespace should succeed, and all tablespace filesystem changes should exist.
Co-authored-by: NTaylor Vesely <tvesely@pivotal.io>
Co-authored-by: NAlexandra Wang <lewang@pivotal.io>
Co-authored-by: NSoumyadeep Chakraborty <sochakraborty@pivotal.io>

* Add remote_regress folder in src/bin/gpfdist/
to test gpfdist which runs on a remote host, such
as a windows server. make installcheck_win can
run them.

* The scripts of start_gpfdist_remote_win
are used to start and stop gpfdist runing on windows.
If you want to test gpfdist on another os, such
as aix, you can create a forlder named
start_gpfdist_remote_aix and add some scripts
to start and stop gpfdist.