Integer overflow occurs without this when copied more than 2^31 rows,
under the `COPY ON SEGMENT` mode.

Errors happen when it is casted to uint64, the type of `processed` in
`CopyStateData`, third-party Postgres driver, which takes it as an
int64, fails out of range.

* docs - resgroup memory_auditor to cat/view/gptoolkit

* cgroup mem usage output - used and limit_granted

* updates for code refactor that was just merged

* type to text

* docs: update cast information for GPDB 5/6

Update cast information. Add information about limited text casts.
See section "Type Casts"

* docs: review comments for GPDB 5/6 updated cast information.

* docs: fix typos in updated CAST info

Authored-by: NTodd Sedano <tsedano@pivotal.io>

To "rebalance" a primary-mirror pair, gprecoverseg -r performs the following steps:

1.  bring down the acting primary
2.  issue a query that triggers the failover
3.  bring up the mirror (gprecoverseg -F)

Currently these 3 steps are happening in close succession. However, there is a chance that between step 2 and step 3, the mirror promotion happens slower than we expect. The implicit assumption here is that the acting mirror has finished transitioning to the primary role before step 3 is performed.

This patch adds a retry in "sort of step 2, definitely before step 3", to ensure a good state before we can bring up the mirror.

Co-authored-by: Jesse Zhang sbjesse@gmail.com
Co-authored-by: David Kimura dkimura@pivotal.io

Since protocols do not belong to a namespace, we do not want to dump
them in table or schema filtered backups. They will only be dumped in a
full backup.
Co-authored-by: NKaren Huddleston <khuddleston@pivotal.io>
Co-authored-by: NChris Hajas <chajas@pivotal.io>

If the column statistics in `pg_statistic` has values with type
different than column type, metadata accessor should not translate the
stats and create a dummy stats instead.

This commit also reorders stats collection from the `pg_statistic` to
align with how analyze generates stats. MCV and Histogram translation is
moved to the end after NDV, nullfraction and column width extraction.
Signed-off-by: NMelanie Plageman <mplageman@pivotal.io>

Co-authored-by: NDavid Sharp <dsharp@pivotal.io>
Co-authored-by: NLarry Hamel <lhamel@pivotal.io>

Consider the following SQL, we expect logging to be turned off for table
`ext_error_logging_off`

```sql
create external table ext_error_logging_off (a int, b int)
    location ('file:///tmp/test.txt') format 'text'
    segment reject limit 100;
\d+ ext_error_logging_off
```
And then in this next case we expect error logging to be turned on for
table `ext_t2`:

```sql
create external table ext_t2 (a int, b int)
    location ('file:///tmp/test.txt') format 'text'
    log errors segment reject limit 100;
\d+ ext_t2
```

Before this patch, we are making two mistakes in handling these external
table DDL:

1. We intend to enable error logging *whenever* the user specifies
`SEGMENT REJECT` clause, completely ignoring whether he or she specifies
`LOG ERRORS` 1. Even then, we make the mistake of implicitly coercing
the OID (an unsigned 32-bit integer) to a bool (which is really just a C
`char`): that means, 255/256 of the time (99.6%) the result is `true`,
and 0.4% of the time we get a `false` instead.

The `OID` to `bool` implicit conversion could have been caught by a
`-Wconversion` GCC/Clang flag. It's most likely a leftover from commit
8f6fe2d6.

This bug manifests itself in the `dsp` regression test mysteriously
failing about once every 200 runs -- with the only diff on a `\d+` of an
external table that should have error logging turned on, but the
returned definition has it turned off.

While working on this we discovered that all of our existing external
tables have both `LOG ERRORS` and `SEGMENT REJECT`, which is why this
bug wasn't caught in the first place.

This patch fixes the issue by properly setting the catalog column
`pg_exttable.logerrors` according to the user input.

While we were at it, we also cleaned up a few dead pieces of code and
made the `dsp` test a bit friendlier to debug.
Co-authored-by: NJesse Zhang <sbjesse@gmail.com>
Co-authored-by: NDavid Kimura <dkimura@pivotal.io>

Fixes greenplum-db/gporca#358

QE readers incorrectly return true for TransactionIdIsCurrentTransactionId()
when passed with an xid that is an aborted subtransaction of current
transaction.  The end effect is wrong results because tuples inserted by the
aborted subtransaction are seen (treated as visible according to MVCC rules) by
a reader.  Current patch fixes the bug by looking up abort status of an XID
from pg_clog.  In a QE writer, just like in upstream PostgreSQL, subtransaction
information is available in CurrentTransactionState (even when subxip cache has
overflown).  This information is not maintained in shared memory, making it
unavailable to a reader.  Readers must resort to a longer route to get the same
information - pg_subtrans and pg_clog.

The patch does not use TransactionIdDidAbort() to check abort status.  The
interface is designed to work with all transaction IDs.  It walks up the
transaction hierarchy to look for an aborted parent if status of the given
transaction is found to be SUB_COMMITTED.  This is a wasted effort when a QE
reader wants to test if its own subtransaction has aborted or not.  A new
interface is introduced to avoid this wasted effort for QE readers.  We choose
to rely on AbortSubTransaction()'s behavior to mark entire subtree under the
aborted subtransaction to be aborted in pg_clog.  A SUB_COMMITTED status in
pg_clog, therefore, allows us to conclude that the subtransaction is not
aborted without having to walk up the hierarchy, provided, the subtransaction
is child of our own transaction.

The test case also needed a fix because the SQL query (insert into select *)
didn't result in a reader gang being created.  The SQL is changed to a join on
a non-distribution column so as to achieve reader gang creation.

Temp tables don't have to be replicated neither crash safe and hence avoid
generating xlog records for them. Heap already avoids, this patch skips for
AO/CO tables as well. Adding new variable `isTempRel` to `BufferedAppend` to
help perform the check for temp tables and skip generating xlog records.

Printable filters are used to produce an expression for the Partition
Selector node that is printed during EXPLAIN to give a hint to the
general nature of the filter used by the node. It is not used by the
executor in any way which actually uses levelEqExpressions,
levelExpressions & residualPredicate instead. These usually contain
completely different set of expressions such as PartBounExpr etc. which
are not printed during EXPLAIN.

Also, with dynamic partition elimination, the partition selector's
printable filter may contain VARs that are not in its subtree and
instead refer to a DynamicTableScan node on the other side of a Join.
This means that it becomes tricky to extract the correct printable
filter expression during DXL to PlStmt translation since that occurs
bottom-up.

Since it is misleading and sometimes incorrect, it's better to remove it
altogether.
Signed-off-by: NSambitesh Dash <sdash@pivotal.io>

*** CID 185522:  Security best practices violations  (STRING_OVERFLOW)
/tmp/build/0e1b53a0/gpdb_src/src/backend/cdb/cdbtm.c: 2486 in gatherRMInDoubtTransactions()

and

*** CID 185520:  Null pointer dereferences  (FORWARD_NULL)
/tmp/build/0e1b53a0/gpdb_src/src/backend/storage/ipc/procarray.c: 2251 in GetSnapshotData()

This condition cannot happen as `GetDistributedSnapshotMaxCount()` doesn't
return 0 for DTX_CONTEXT_QD_DISTRIBUTED_CAPABLE and hence `inProgressXidArray`
will always be initialzed. hence marked as ignore in coverity but still worth
adding Assert for the same.

Because the pg_dump handle the implicit sequence (serial column type)
and explicit sequence differently in different gpdb version, so we
need to detect the sequences which not included in the table dump sql,
then we should create them firstly. And also for all dependent sequence,
setval() to source value after data transferred.
Signed-off-by: NMing Li <mli@pivotal.io>

The persistent field of LOCK struct may cause confusion. Its meaning is the lock can only
be released after the transaction ends.  This commit renames this field as holdTillEndXact
and changes some related function's name. Some comments are also added.

Prior to this commit, gpaddmirrors was missing two bits of work which
were previously done by gpinitsystem/gpcreateseg. When adding mirrors to
a cluster, the pg_hba.conf on primary segments has to be modified to
allow replication connections, e.g. for pgbasebackup. And after the
mirrors are built, the catalog has to be updated to reflect the new
cluster config.
Co-authored-by: NNadeem Ghani <nghani@pivotal.io>
Co-authored-by: NJim Doty <jdoty@pivotal.io>

Co-authored-by: NJason Vigil <jvigil@pivotal.io>
Co-authored-by: NKris Macoskey <kmacoskey@pivotal.io>

Co-authored-by: NJason Vigil <jvigil@pivotal.io>
Co-authored-by: NTrevor Yacovone <tyacovone@pivotal.io>

Code had these two variables (GUCs), serving same purpose. GpIdentity.segindex
is set to content-id, based on command line argument at start-up and inherited
by all processes from postmaster. Whereas Gp_segment, was session level guc only
set for backends, by dispatching the same from QD. So. essentially Gp_segment
was not available and had incorrect value in auxiliary processes.

Hence replaced all usages with GpIdentity.segindex. As a side effect log files
now have correct value reported for segment number (content-id) in each and
every line of file, irrespective of which process generated the log message.

Discussion:
https://groups.google.com/a/greenplum.org/forum/#!msg/gpdb-dev/Yr8-LZIiNfA/ob4KLgmkAQAJ

This patch helps cut-down 8 secs out of 30 secs (for single node gpinitsystem as
part of gpdemo), mostly more for multinode setups.

- Removes explicit 2 sec and 1 sleep.
- Removes explicit call to CHECKPOINT, not required.
- Avoid starting primaries after initdb. As they were started and then shutdown
  pretty soon to start back cluster with required cmdline arguments.
- Add -w to mirror start.

Still more room for improvement and speedup, this just gets us started.

pycrypto version is now identical with the version in
python-dependencies.txt
Authored-by: NTodd Sedano <tsedano@pivotal.io>

Parse version is now identical with the version in
python-dependencies.txt
Authored-by: NTodd Sedano <tsedano@pivotal.io>

We should pass binary_swap_gpdb_centos6 instead of gpdb_src_binary_swap.

We used to implement the memory auditor feature differently on master
and 5X, on master the attribute is stored in pg_resgroup while on 5X
it's stored in pg_resgroupcapability.  This increases the maintenance
effort significantly.  So we refactor this feature on master to minimize
the difference between these two branches.

- Revert "resgroup: fix an access to uninitialized address."
  This reverts commit 56c20709.
- Revert "ic: Mark binary_swap_gpdb as optional input for resgroup jobs."
  This reverts commit 9b3d0cfc.
- Revert "resgroup: fix a boot failure when cgroup is not mounted."
  This reverts commit 4c8f28b0.
- Revert "resgroup: backward compatibility for memory auditor"
  This reverts commit f2f86174.
- Revert "Show memory statistics for cgroup audited resource group."
  This reverts commit d5fb628f.
- Revert "Fix resource group test failure."
  This reverts commit 78b885ec.
- Revert "Support cgroup memory auditor for resource group."
  This reverts commit 6b3d0f66.
- Apply "resgroup: backward compatibility for memory auditor"
  This cherry picks commit 23cd8b1e
- Apply "ic: Mark binary_swap_gpdb as optional input for resgroup jobs."
  This Cherry picks commit c86652e6
- Apply "resgroup: fix an access to uninitialized address."
  This Cherry picks commit b257b344

This guc was removed by accident in
5cc0ec50.

In case we remove the gucs by accident in the future, we added a
testcase to check for the existence of the resgroup gucs.

Same transaction truncate performs unsafe truncation, hence adding test to cover
for the same.

Upstream commit cab9a065 introduced an optimization to truncate tables
in scenarios that permit "unsafe" operations where we don't have to
churn on the relfilenode for the underlying tables. AO table got a free
ride but for the wrong reason.

This patch teaches heap_truncate_one_rel() to perform the unsafe /
optimal truncation on AO tables. This allows us to converge the callers
back to how they look like in Postgres 9.0.

Specifically, we're now able to inline TruncateRelfiles() back into
ExecuteTruncate() .

One caveat introduced by this patch though, is the "optimal" / unsafe
truncation of an AO table can potentially leak some disk space: we are
not performing a real file-level truncate, merely seeking back to offset
0 -- because the aoseg auxiliary table is truncated -- on the next
write, therefore the space after the EOF mark is wasted in some sense.
Co-authored-by: NJesse Zhang <sbjesse@gmail.com>

Missed one reference

[ci skip]
Authored-by: NTodd Sedano <tsedano@pivotal.io>

[ci skip]
Authored-by: NTodd Sedano <tsedano@pivotal.io>

There were scenarios where adding a new partition to a partition table would
cause a negative or duplicate partition rule order (parruleord) value to show
up in the pg_partition_rule catalog table.

1. Negative parruleord values could show up during parruleord gap closing when
   the new partition is inserted above a parruleord gap.
2. Negative parruleord values could show up when the max number of partitions
   for that level has been reached (32767), and there is an attempt to add a
   new partition that would have been the highest ranked partition in that
   partition's partition range.
3. Duplicate parruleord values could show up when the max number of partitions
   for that level has been reached (32767), and there is an attempt to add a
   new partition that would have been inserted between the partition table's
   sequence of parruleord values.
Co-authored-by: NDavid Kimura <dkimura@pivotal.io>

To avoid disk space spamming while running tests and also save some time, patch
intends to avoid core file generation for intentional PANICs caused by tests.

Using `setrlimit()` to achieve the same is output of discussion with Jacob
Champion and Asim Praveen. Alternative option considered was calling
`quickdie()` instead of PANIC, but that suffers from not providing feedback
PANIC string used by test to validate the reason for PANIC.

GNU readline defines the return value of write_history() as "zero if OK,
else an errno code".  libedit's version of that function used to have a
different definition (to wit, "-1 if error, else the number of lines
written to the file").  We tried to work around that by checking whether
errno had become nonzero, but this method has never been kosher according
to the published API of either library.  It's reportedly completely broken
in recent Ubuntu releases: psql bleats about "No such file or directory"
when saving ~/.psql_history, even though the write worked fine.

However, libedit has been following the readline definition since somewhere
around 2006, so it seems all right to finally break compatibility with
ancient libedit releases and trust that the return value is what readline
specifies.  (I'm not sure when the various Linux distributions incorporated
this fix, but I did find that OS X has been shipping fixed versions since
10.5/Leopard.)

If anyone is still using such an ancient libedit, they will find that psql
complains it can't write ~/.psql_history at exit, even when the file was
written correctly.  This is no worse than the behavior we're fixing for
current releases.

Back-patch to all supported branches.

(cherry picked from commit df9ebf1e)
Fixes #4437.

The test can fail if PANIC happens before insert has even reached the intended
point of finishing commit on master. In this case select rightfully returns zero
rows compared to expected 1 row.