Need to close pipe files even if COPY PROGRAME already exit before parent process ask child process to exit, otherwise it will lead file description leak in a long time running.

When building without assertions, any storage allocated for an
assertion will generate a compiler warning for unused variable
or function. Wrap all such locations in the USE_ASSERT_CHECKING
macro to remove warnings, or remove the need for storage in the
first place.

The inline functions were single statement functions written on a
single line, which breaks the unittest mocker code when preprocessor
directives are added after these functions. The reason for this is
that the mocker code hunts for an ending brace on its own line. As a
later commit will introduce said preprocessor directive, it's time
to bite the bullet and fix this.

listen_addresses in postgresql.conf doesn't taken effect now, backend and
postmaster are listening on all addresses. From the point of security, we
should be able to let user specify the listen address.

oldcontext may not be set here, and if it is we will be in that
context already.

The new -Wformat-overflow warning in GCC7 fails to understand that
the buffer is in fact large enough, likely because it counts on the
integers possibly being negative. Move to using snprintf() instead
since that accounts for buffer size, and is really what we should
have done in the first place.  Also touch up the consumer of the
GP_CSVOPT buffer even though it's using a buffer oversized enough
to probably not cause that same warning.

They are ignored by gpdiff, but it's nice to keep them up-to-date anyway,
because if a test fails, these differences alre also printed in
regression.diffs, even though they're not causing the failure.

There are some more CONTEXTs that in more complicated cases that are not
always the same from run to run, but this is a good start.

If an error happens in the prepare phase of two-phase commit, relay the
original error back to the client, instead of the fairly opaque
"Abort [Prepared]' broadcast failed to one or more segments" message you
got previously. A lot of things happen during the prepare phase that
can legitimately fail, like checking deferred constraints, like in the
'constraints' regression test. But even without that, there can be
triggers, ON COMMIT actions, etc., any of which can fail.

This commit consists of several parts:

* Pass 'true' for the 'raiseError' argument when dispatching the prepare
  dtx command in doPrepareTransaction(), so that the error is emitted to
  the client.

* Bubble up an ErrorData struct, with as many fields intact as possible,
  to the caller,  when dispatching a dtx command. (Instead of constructing
  a message in a StringInfo). So that we can re-throw the message to
  the client, with its original formatting.

* Don't throw an error in performDtxProtocolCommand(), if we try to abort
  a prepared transaction that doesn't exist. That is business-as-usual,
  if a transaction throws an error before finishing the prepare phase.

* Suppress the "NOTICE: Releasing segworker groups to retry broadcast."
  message, when aborting a prepared transaction.

Put together, the effect is if an error happens during prepare phase, the
client receives a message that is largely indistinguishable from the
message you'd get if the same failure happened while running a normal
statement.

Fixes github issue #4530.

I overlooked that these errors were thrown at wrong place, after COMMIT,
because of the difference in the error message.

Commit ca4c29d4 removed the error
deferring from EXPLAIN such that it attempted to return results
even in case of errors, but missed the DXL case. This removes the
TRY/CATCH wraps from DXL EXPLAIN to match the rest of the EXPLAIN
code.

Also remove the last (found) callsite that expects that a palloc
for a statsbuffer can have returned NULL into an Assertion.

When generating DXL output in EXPLAIN {ANALYZE}, make sure to catch
any exceptions generated on the C++ side to avoid a server crash on
queries where ORCA fail to generate a plan. A better job of catching
the various different possible exceptions and generating nice error
messages can be made, this closes the current hole of trivial core
dumps on for example this: EXPLAIN (dxl) SELECT * FROM pg_class;

Install go dep by default, and remove ~/go from GOPATH.

This test panics the segment, so muts make sure segment is recovered
fully before moving ahead. Commit
2fea28fb tried to fix a failure due to
it by moving gp_toolkit tests ahead of it but that's not right
solution as some other test can hit the condition or within this test
itself condition can be hit. Hence moving it to isolation2 and adding
explicit check to validate segment has recovered before moving ahead.

Different queries in the same session may be in different resource queues.
The resource queue field of stat cache in local hash should be cleaned
after the stat message has been sent out.

This commit fix the Github issue #4582 .

Co-authored-by: NDavid Kimura <dkimura@pivotal.io>

* docs - dblink works w/ dbs of same greenplum major version

* cannot dblink to postgresql, remove ref

Co-authored-by: NDavid Kimura <dkimura@pivotal.io>

FTS makes updates to cluster configuration metadata tables.  These
updates are made within a transaction.  If an error is encountered,
the transaction must be aborted and all locks must be released before
FTS process exits.

A new ICW test is added to simulate error during configuration update.
Co-authored-by: NDavid Kimura <dkimura@pivotal.io>

Test added for ORCA version v2.55.4
Signed-off-by: NHaisheng Yuan <hyuan@pivotal.io>

Signed-off-by: NBhuvnesh Chaudhary <bchaudhary@pivotal.io>

* docs - note OOM query failure when resource groups active

* address comment from simon; add info to rq/rg differences table

* edits from david

Older platforms are behind in certificates with the recent change in
upstream github for the allowable TLS versions when clients connect.

- if the data file contains "\N" as the delimiter, it would not be
recognized properly by gpload
- root cause: gpload replace the quote in nullas option as well as
replace '\' as '\\'
- solution: add quote_no_slash function to handle nullas option

*  Add PG_BASEBACKUP_FIXME:if there are new tablespaces, gpinitstandby fails.
   pg_basebackup can't copy tablespace data to standby, besause the tablespace
   directory is not empty on standby.
*  Check standby tablespace directory.
*  Replace function cleanupFilespaces with function cleanup_tablespace.

  Author: Max Yang <myang@pivotal.io>
  Author: Xiaoran Wang <xiwang@pivotal.io>

* resgroup: make cgroup memsw.limit_in_bytes optional.
* resgroup: retry proc migration for rmdir to succeed.
* resgroup: add delay in a testcase.
* resgroup: use correct log level in cgroup ops.

This commit moves some cleanup code to finally blocks to prevent
gpdeletesystem hanging on error cases.
Co-authored-by: NJamie McAtamney <jmcatamney@pivotal.io>
Co-authored-by: NNadeem Ghani <nghani@pivotal.io>

In minirepro, the following query is generated:
```
DELETE FROM pg_statistic WHERE starelid='tpcds.catalog_returns_1_prt_27'::regclass AND staattnum=22;
INSERT INTO pg_statistic VALUES ......
```
It only has existing stats entries for catalog tables, so for non-catalog
table, the DELETE command is useless, and slows down the speed of loading
minirepro.

We should NOT generate DELETE query for non-catalog tables.
Signed-off-by: NEkta Khanna <ekhanna@pivotal.io>

gp_toolkit tests runs a query which checks if there are any GUCs
which have different value on the cluster node. However, test
udf_exception_blocks_panic_scenarios is executed before it and it
injects PANIC which causes recovery and it takes some time for all
the GUCs to be in consistent state between the segments. But by that
time gp_toolkit identifies difference and fails, so moving gp_toolkit
test above it.

GPORCA uses NOT_EXIST_SUBLINK to implement co-related left anti semijoin.
Therefore we still need this in the executor for GPORCA plans.

Example:

```sql
CREATE TABLE csq_r(a int) distributed by (a);
INSERT INTO csq_r VALUES (1);
CREATE OR REPLACE FUNCTION csq_f(a int) RETURNS int AS $$ select $1 $$ LANGUAGE SQL CONTAINS SQL;
explain SELECT * FROM csq_r WHERE not exists (SELECT * FROM csq_f(csq_r.a));
Physical plan:
+--CPhysicalMotionGather(master)   rows:1   width:34  rebinds:1   cost:882688.037301   origin: [Grp:7, GrpExpr:14]
   +--CPhysicalCorrelatedLeftAntiSemiNLJoin("" (8))   rows:1   width:34  rebinds:1   cost:882688.037287   origin: [Grp:7, GrpExpr:13]
      |--CPhysicalTableScan "csq_r" ("csq_r")   rows:1   width:34  rebinds:1   cost:431.000019   origin: [Grp:0, GrpExpr:1]
      |--CPhysicalComputeScalar   rows:1   width:1  rebinds:1   cost:0.000002   origin: [Grp:5, GrpExpr:1]
      |  |--CPhysicalConstTableGet Columns: ["" (8)] Values: [(1)]   rows:1   width:1  rebinds:1   cost:0.000001   origin: [Grp:1, GrpExpr:1]
      |  +--CScalarProjectList   origin: [Grp:4, GrpExpr:0]
      |     +--CScalarProjectElement "csq_f" (9)   origin: [Grp:3, GrpExpr:0]
      |        +--CScalarIdent "a" (0)   origin: [Grp:2, GrpExpr:0]
      +--CScalarConst (1)   origin: [Grp:8, GrpExpr:0]
",
                                   QUERY PLAN
---------------------------------------------------------------------------------
 Gather Motion 3:1  (slice1; segments: 3)  (cost=0.00..882688.04 rows=1 width=4)
   Output: a
   ->  Table Scan on public.csq_r  (cost=0.00..882688.04 rows=1 width=4)
         Output: a
         Filter: (SubPlan 1)
         SubPlan 1  (slice1; segments: 3)
           ->  Result  (cost=0.00..0.00 rows=1 width=1)
                 Output:
                 ->  Result  (cost=0.00..0.00 rows=1 width=1)
                       Output: $0,
                       ->  Result  (cost=0.00..0.00 rows=1 width=1)
                             Output: true
 Optimizer: PQO version 2.55.2
(13 rows)
```

Prior to this commit, on calling `RESET`/`RESET ALL` command, the value of guc
was not getting updated on all the slices(already spun up reader slices).
This commit fixes the issue.
Signed-off-by: NEkta Khanna <ekhanna@pivotal.io>

There is a scenario during AO/CO delete or update where the first row number
obtained is negative. The error is caused when the first row number in the
aovisimap of an AO/CO table exceeds int max. There is currently a bug where
the first row number is retrieved from the tuple as an int, but the first row
number is an int64. We fixed this by retrieving as an int64.
Co-authored-by: NJimmy Yih <jyih@pivotal.io>
Co-authored-by: NDavid Kimura <dkimura@pivotal.io>

These underlying issues  where resolved in the following PRs that were
merged. The fixmes where not removed.

https://github.com/greenplum-db/gpdb/pull/4174/files
https://github.com/greenplum-db/gporca/pull/263