Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sat, Jul 10, 2021 at 5:19 PM John Naylor wrote: > Side note for future reference: While the feature named in the CF entry has > been rejected, the remaining 0001 patch currently proposed no longer matches > the title, or category. It is possible within the CF app, and helpful, to > rename the entry when the scope changes. > > The proposed patch in the CF for incremental view maintenance [1] does some > refactoring of its own in implementing the feature. I don't think it makes > sense to commit a refactoring that conflicts with that, while not necessarily > making life easier for that feature. Incremental view maintenance is highly > desirable, so I don't want to put up unnecessary roadblocks. Thanks. I'm okay to close the CF entry(https://commitfest.postgresql.org/33/2928/) and stop this thread. Regards, Bharath Rupireddy.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Tue, Mar 16, 2021 at 8:13 AM Bharath Rupireddy < bharath.rupireddyforpostg...@gmail.com> wrote: > > On Tue, Mar 16, 2021 at 1:15 AM Tom Lane wrote: > > I don't really see that this feature buys us anything you can't > > get by explaining the view's query, so I think we're better advised > > to keep our options open about how REFRESH might be implemented > > in future. > > That makes sense to me. Thanks for the comments. I'm fine to withdraw the patch. > > I would like to see if the 0001 patch(attaching here) will be useful > at all. It just splits up the existing ExecRefreshMatView into a few > functions to make the code readable. I'm okay to withdraw it if no one > agrees. Side note for future reference: While the feature named in the CF entry has been rejected, the remaining 0001 patch currently proposed no longer matches the title, or category. It is possible within the CF app, and helpful, to rename the entry when the scope changes. The proposed patch in the CF for incremental view maintenance [1] does some refactoring of its own in implementing the feature. I don't think it makes sense to commit a refactoring that conflicts with that, while not necessarily making life easier for that feature. Incremental view maintenance is highly desirable, so I don't want to put up unnecessary roadblocks. [1] https://commitfest.postgresql.org/33/2138/ -- John Naylor EDB: http://www.enterprisedb.com
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Tue, 16 Mar 2021 at 20:13, Bharath Rupireddy wrote: > On Tue, Mar 16, 2021 at 1:15 AM Tom Lane wrote: >> >> [ Sorry for not looking at this thread sooner ] >> >> Bharath Rupireddy writes: >> > Currently, $subject is not allowed. We do plan the mat view query >> > before every refresh. I propose to show the explain/explain analyze of >> > the select part of the mat view in case of Refresh Mat View(RMV). >> >> TBH, I think we should reject this. The problem with it is that it >> binds us to the assumption that REFRESH MATERIALIZED VIEW has an >> explainable plan. There are various people poking at ideas like >> incremental matview updates, which might rely on some implementation >> that doesn't exactly equate to a SQL query. Incremental updates are >> hard enough already; they'll be even harder if they also have to >> maintain compatibility with a pre-existing EXPLAIN behavior. >> >> I don't really see that this feature buys us anything you can't >> get by explaining the view's query, so I think we're better advised >> to keep our options open about how REFRESH might be implemented >> in future. > > That makes sense to me. Thanks for the comments. I'm fine to withdraw the > patch. > > I would like to see if the 0001 patch(attaching here) will be useful > at all. It just splits up the existing ExecRefreshMatView into a few > functions to make the code readable. +1. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Tue, Mar 16, 2021 at 1:15 AM Tom Lane wrote:
>
> [ Sorry for not looking at this thread sooner ]
>
> Bharath Rupireddy writes:
> > Currently, $subject is not allowed. We do plan the mat view query
> > before every refresh. I propose to show the explain/explain analyze of
> > the select part of the mat view in case of Refresh Mat View(RMV).
>
> TBH, I think we should reject this. The problem with it is that it
> binds us to the assumption that REFRESH MATERIALIZED VIEW has an
> explainable plan. There are various people poking at ideas like
> incremental matview updates, which might rely on some implementation
> that doesn't exactly equate to a SQL query. Incremental updates are
> hard enough already; they'll be even harder if they also have to
> maintain compatibility with a pre-existing EXPLAIN behavior.
>
> I don't really see that this feature buys us anything you can't
> get by explaining the view's query, so I think we're better advised
> to keep our options open about how REFRESH might be implemented
> in future.
That makes sense to me. Thanks for the comments. I'm fine to withdraw the patch.
I would like to see if the 0001 patch(attaching here) will be useful
at all. It just splits up the existing ExecRefreshMatView into a few
functions to make the code readable. I'm okay to withdraw it if no one
agrees.
With Regards,
Bharath Rupireddy.
EnterpriseDB: http://www.enterprisedb.com
From 26970ffd33e67324609a03f0f61eeb6406216a7f Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy
Date: Fri, 5 Mar 2021 15:47:12 +0530
Subject: [PATCH v7 1/2] Rearrange Refresh Mat View Code
Currently, the function ExecRefreshMatView in matview.c is having
many lines of code which is not at all good from readability and
maintainability perspectives. This patch adds few functions and
moves the code from ExecRefreshMatView to them making the code
look better.
---
src/backend/commands/matview.c | 449 -
1 file changed, 269 insertions(+), 180 deletions(-)
diff --git a/src/backend/commands/matview.c b/src/backend/commands/matview.c
index c5c25ce11d..18e18fa627 100644
--- a/src/backend/commands/matview.c
+++ b/src/backend/commands/matview.c
@@ -64,7 +64,7 @@ static void transientrel_startup(DestReceiver *self, int operation, TupleDesc ty
static bool transientrel_receive(TupleTableSlot *slot, DestReceiver *self);
static void transientrel_shutdown(DestReceiver *self);
static void transientrel_destroy(DestReceiver *self);
-static uint64 refresh_matview_datafill(DestReceiver *dest, Query *query,
+static uint64 refresh_matview_datafill(Oid OIDNewHeap, Query *query,
const char *queryString);
static char *make_temptable_name_n(char *tempname, int n);
static void refresh_by_match_merge(Oid matviewOid, Oid tempOid, Oid relowner,
@@ -73,6 +73,16 @@ static void refresh_by_heap_swap(Oid matviewOid, Oid OIDNewHeap, char relpersist
static bool is_usable_unique_index(Relation indexRel);
static void OpenMatViewIncrementalMaintenance(void);
static void CloseMatViewIncrementalMaintenance(void);
+static Query *get_matview_query(RefreshMatViewStmt *stmt, Relation *rel,
+Oid *objectId);
+static Query *rewrite_refresh_matview_query(Query *dataQuery);
+static Oid get_new_heap_oid(RefreshMatViewStmt *stmt, Relation matviewRel,
+ Oid matviewOid, char *relpersistence);
+static void match_matview_with_new_data(RefreshMatViewStmt *stmt,
+ Relation matviewRel, Oid matviewOid,
+ Oid OIDNewHeap, Oid relowner,
+ int save_sec_context,
+ char relpersistence, uint64 processed);
/*
* SetMatViewPopulatedState
@@ -140,127 +150,18 @@ ExecRefreshMatView(RefreshMatViewStmt *stmt, const char *queryString,
{
Oid matviewOid;
Relation matviewRel;
- RewriteRule *rule;
- List *actions;
Query *dataQuery;
- Oid tableSpace;
- Oid relowner;
Oid OIDNewHeap;
- DestReceiver *dest;
uint64 processed = 0;
- bool concurrent;
- LOCKMODE lockmode;
+ Oid relowner;
char relpersistence;
Oid save_userid;
int save_sec_context;
int save_nestlevel;
ObjectAddress address;
- /* Determine strength of lock needed. */
- concurrent = stmt->concurrent;
- lockmode = concurrent ? ExclusiveLock : AccessExclusiveLock;
-
- /*
- * Get a lock until end of transaction.
- */
- matviewOid = RangeVarGetRelidExtended(stmt->relation,
- lockmode, 0,
- RangeVarCallbackOwnsTable, NULL);
- matviewRel = table_open(matviewOid, NoLock);
-
- /* Make sure it is a materialized view. */
- if (matviewRel->rd_rel->relkind != RELKIND_MATVIEW)
- ereport(ERROR,
-(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
- errmsg("\"%s\" is not a materialized view",
- RelationGetRelationName(matviewRel;
-
- /* Check that CONCURRENTLY is not specified if not populated. */
- if (concurrent && !RelationIsPopulated(matviewRel))
- ereport(ERROR,
-(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
- errmsg("CONCURRENTLY cannot be used when the
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
[ Sorry for not looking at this thread sooner ] Bharath Rupireddy writes: > Currently, $subject is not allowed. We do plan the mat view query > before every refresh. I propose to show the explain/explain analyze of > the select part of the mat view in case of Refresh Mat View(RMV). TBH, I think we should reject this. The problem with it is that it binds us to the assumption that REFRESH MATERIALIZED VIEW has an explainable plan. There are various people poking at ideas like incremental matview updates, which might rely on some implementation that doesn't exactly equate to a SQL query. Incremental updates are hard enough already; they'll be even harder if they also have to maintain compatibility with a pre-existing EXPLAIN behavior. I don't really see that this feature buys us anything you can't get by explaining the view's query, so I think we're better advised to keep our options open about how REFRESH might be implemented in future. regards, tom lane
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sat, Mar 13, 2021 at 7:00 AM Japin Li wrote: > > On Mon, 08 Mar 2021 at 12:28, Bharath Rupireddy > wrote: > > On Sun, Mar 7, 2021 at 10:13 PM Zhihong Yu wrote: > >> Hi, > >> > >> +* EXPLAIN ANALYZE CREATE TABLE AS or REFRESH MATERIALIZED VIEW > >> +* WITH NO DATA is weird. > >> > >> Maybe it is clearer to spell out WITH NO DATA for both statements, instead > >> of sharing it. > > > > Done that way. > > > >> - if (!stmt->skipData) > >> + if (!stmt->skipData && !explainInfo) > >> ... > >> + else if (explainInfo) > >> > >> It would be cleaner to put the 'if (explainInfo)' as the first check. That > >> way, the check for skipData can be simplified. > > > > Changed. > > > > Thanks for review comments. Attaching v7 patch set with changes only > > in 0002 patch. Please have a look. > > > > The v7 patch looks good to me, and there is no other advice, so I change > the status to "Ready for Committer". Thanks for the review. With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Mon, 08 Mar 2021 at 12:28, Bharath Rupireddy wrote: > On Sun, Mar 7, 2021 at 10:13 PM Zhihong Yu wrote: >> Hi, >> >> +* EXPLAIN ANALYZE CREATE TABLE AS or REFRESH MATERIALIZED VIEW >> +* WITH NO DATA is weird. >> >> Maybe it is clearer to spell out WITH NO DATA for both statements, instead >> of sharing it. > > Done that way. > >> - if (!stmt->skipData) >> + if (!stmt->skipData && !explainInfo) >> ... >> + else if (explainInfo) >> >> It would be cleaner to put the 'if (explainInfo)' as the first check. That >> way, the check for skipData can be simplified. > > Changed. > > Thanks for review comments. Attaching v7 patch set with changes only > in 0002 patch. Please have a look. > The v7 patch looks good to me, and there is no other advice, so I change the status to "Ready for Committer". -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sun, Mar 7, 2021 at 10:13 PM Zhihong Yu wrote: > Hi, > > +* EXPLAIN ANALYZE CREATE TABLE AS or REFRESH MATERIALIZED VIEW > +* WITH NO DATA is weird. > > Maybe it is clearer to spell out WITH NO DATA for both statements, instead of > sharing it. Done that way. > - if (!stmt->skipData) > + if (!stmt->skipData && !explainInfo) > ... > + else if (explainInfo) > > It would be cleaner to put the 'if (explainInfo)' as the first check. That > way, the check for skipData can be simplified. Changed. Thanks for review comments. Attaching v7 patch set with changes only in 0002 patch. Please have a look. With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com v7-0001-Rearrange-Refresh-Mat-View-Code.patch Description: Binary data v7-0002-EXPLAIN-EXPLAIN-ANALYZE-REFRESH-MATERIALIZED-VIEW.patch Description: Binary data
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
Hi, +* EXPLAIN ANALYZE CREATE TABLE AS or REFRESH MATERIALIZED VIEW +* WITH NO DATA is weird. Maybe it is clearer to spell out WITH NO DATA for both statements, instead of sharing it. - if (!stmt->skipData) + if (!stmt->skipData && !explainInfo) ... + else if (explainInfo) It would be cleaner to put the 'if (explainInfo)' as the first check. That way, the check for skipData can be simplified. Cheers On Sun, Mar 7, 2021 at 1:34 AM Bharath Rupireddy < bharath.rupireddyforpostg...@gmail.com> wrote: > On Sun, Mar 7, 2021 at 12:13 PM Japin Li wrote: > > > > On Sun, 07 Mar 2021 at 14:25, Bharath Rupireddy < > bharath.rupireddyforpostg...@gmail.com> wrote: > > > On Sun, Mar 7, 2021 at 11:49 AM Japin Li wrote: > > >> > > >> On Fri, 05 Mar 2021 at 19:48, Bharath Rupireddy < > bharath.rupireddyforpostg...@gmail.com> wrote: > > >> > Attaching v5 patch set for further review. > > >> > > > >> > > >> The v5 patch looks good to me, if there is no objection, I'll change > the > > >> cf status to "Ready for Committer" in few days. > > > > > > Thanks for the review. > > > > > > As I mentioned upthread, I have 2 open points: > > > 1) In the patch I have added a new mat view info parameter to > > > ExplainOneQuery(), do we also need to add it to > > > ExplainOneQuery_hook_type? IMO, we should not (for now), because this > > > would create a backward compatibility issue. > > > > Sorry, I do not know how PostgreSQL handle the backward compatibility > issue. > > Is there a guideline? > > I'm not aware of any guidelines as such, but we usually avoid any > changes to existing API, adding/making changes to system catalogs and > so on. > > > > 2) Do we document (under respective command pages or somewhere else) > > > that we allow explain/explain analyze for a command? > > > > > > > IMO, we can add a new page to list the commands that can be > explain/explain analyze, > > since it's clear for users. > > We are listing all the supported commands in explain.sgml, so added > the CREATE MATERIALIZED VIEW(it's missing even though it's supported > prior to this patch) and REFRESH MATERIALIZED VIEW there. > > Attaching v6 patch set. Please have a look. > > With Regards, > Bharath Rupireddy. > EnterpriseDB: http://www.enterprisedb.com >
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sun, 07 Mar 2021 at 17:33, Bharath Rupireddy wrote: > On Sun, Mar 7, 2021 at 12:13 PM Japin Li wrote: >> >> On Sun, 07 Mar 2021 at 14:25, Bharath Rupireddy >> wrote: >> > On Sun, Mar 7, 2021 at 11:49 AM Japin Li wrote: >> >> >> >> On Fri, 05 Mar 2021 at 19:48, Bharath Rupireddy >> >> wrote: >> >> > Attaching v5 patch set for further review. >> >> > >> >> >> >> The v5 patch looks good to me, if there is no objection, I'll change the >> >> cf status to "Ready for Committer" in few days. >> > >> > Thanks for the review. >> > >> > As I mentioned upthread, I have 2 open points: >> > 1) In the patch I have added a new mat view info parameter to >> > ExplainOneQuery(), do we also need to add it to >> > ExplainOneQuery_hook_type? IMO, we should not (for now), because this >> > would create a backward compatibility issue. >> >> Sorry, I do not know how PostgreSQL handle the backward compatibility issue. >> Is there a guideline? > > I'm not aware of any guidelines as such, but we usually avoid any > changes to existing API, adding/making changes to system catalogs and > so on. > Thanks for explaining, I'd be inclined keep it backward compatibility. >> > 2) Do we document (under respective command pages or somewhere else) >> > that we allow explain/explain analyze for a command? >> > >> >> IMO, we can add a new page to list the commands that can be explain/explain >> analyze, >> since it's clear for users. > > We are listing all the supported commands in explain.sgml, so added > the CREATE MATERIALIZED VIEW(it's missing even though it's supported > prior to this patch) and REFRESH MATERIALIZED VIEW there. > > Attaching v6 patch set. Please have a look. > LGTM. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
D VIEW is in a portal or plpgsql function and is executed
+ * repeatedly. (See also the same hack in EXPLAIN and PREPARE.)
+ */
+static Query *
+rewrite_refresh_matview_query(Query *dataQuery)
+{
+ List *rewritten;
+ Query *copied_query;
+
+ /* Lock and rewrite, using a copy to preserve the original query. */
+ copied_query = copyObject(dataQuery);
+ AcquireRewriteLocks(copied_query, true, false);
+ rewritten = QueryRewrite(copied_query);
+
+ /* SELECT should never rewrite to more or less than one SELECT query */
+ if (list_length(rewritten) != 1)
+ elog(ERROR, "unexpected rewrite result for REFRESH MATERIALIZED VIEW");
+ dataQuery = (Query *) linitial(rewritten);
+
+ return dataQuery;
+}
--
2.25.1
From 08906356cbbd1fe1231d633847552290e9155ec5 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy
Date: Sun, 7 Mar 2021 14:57:24 +0530
Subject: [PATCH v6] EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
Currently, explain/explain analyze refresh materialized view(RMV)
is not allowed. We do plan the materialized view query before every
refresh. I propose to show the explain/explain analyze of the
select part of the materialized view. It will be useful for the
user to know what exactly is being planned and executed as part of
RMV. Please note that we already have explain/explain analyze
CTAS/Create Mat View(CMV), where we show the explain/explain analyze
of the select part. This proposal will do the same thing.
The behaviour can be like this:
EXPLAIN REFRESH MATERIALIZED VIEW mv1; --> will not refresh the mat
view, but shows the select part's plan of mat view.
EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW mv1; --> will refresh the
mat view and shows the select part's plan of mat view.
---
doc/src/sgml/ref/explain.sgml | 2 +
src/backend/commands/explain.c| 50 +++-
src/backend/commands/matview.c| 36 +---
src/backend/commands/prepare.c| 3 +-
src/backend/tcop/utility.c| 3 +-
src/include/commands/explain.h| 30 +-
src/include/commands/matview.h| 8 ++-
src/test/regress/expected/matview.out | 83 +++
src/test/regress/sql/matview.sql | 27 +
9 files changed, 217 insertions(+), 25 deletions(-)
diff --git a/doc/src/sgml/ref/explain.sgml b/doc/src/sgml/ref/explain.sgml
index c4512332a0..a301b6b53f 100644
--- a/doc/src/sgml/ref/explain.sgml
+++ b/doc/src/sgml/ref/explain.sgml
@@ -95,6 +95,8 @@ EXPLAIN [ ANALYZE ] [ VERBOSE ] statementEXPLAIN ANALYZE on an
INSERT, UPDATE,
DELETE, CREATE TABLE AS,
+CREATE MATERIALIZED VIEW,
+REFRESH MATERIALIZED VIEW,
or EXECUTE statement
without letting the command affect your data, use this approach:
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index afc45429ba..ac15931a41 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -17,6 +17,7 @@
#include "catalog/pg_type.h"
#include "commands/createas.h"
#include "commands/defrem.h"
+#include "commands/matview.h"
#include "commands/prepare.h"
#include "executor/nodeHash.h"
#include "foreign/fdwapi.h"
@@ -53,10 +54,6 @@ explain_get_index_name_hook_type explain_get_index_name_hook = NULL;
#define X_CLOSE_IMMEDIATE 2
#define X_NOWHITESPACE 4
-static void ExplainOneQuery(Query *query, int cursorOptions,
- IntoClause *into, ExplainState *es,
- const char *queryString, ParamListInfo params,
- QueryEnvironment *queryEnv);
static void ExplainPrintJIT(ExplainState *es, int jit_flags,
JitInstrumentation *ji);
static void report_triggers(ResultRelInfo *rInfo, bool show_relname,
@@ -274,7 +271,8 @@ ExplainQuery(ParseState *pstate, ExplainStmt *stmt,
{
ExplainOneQuery(lfirst_node(Query, l),
CURSOR_OPT_PARALLEL_OK, NULL, es,
- pstate->p_sourcetext, params, pstate->p_queryEnv);
+ pstate->p_sourcetext, params, pstate->p_queryEnv,
+ NULL);
/* Separate plans with an appropriate separator */
if (lnext(rewritten, l) != NULL)
@@ -357,11 +355,11 @@ ExplainResultDesc(ExplainStmt *stmt)
*
* "into" is NULL unless we are explaining the contents of a CreateTableAsStmt.
*/
-static void
+void
ExplainOneQuery(Query *query, int cursorOptions,
IntoClause *into, ExplainState *es,
const char *queryString, ParamListInfo params,
-QueryEnvironment *queryEnv)
+QueryEnvironment *queryEnv, RefreshMatViewInfo *matviewInfo)
{
/* planner will not cope with utility statements */
if (query->commandType == CMD_UTILITY)
@@ -402,7 +400,8 @@ ExplainOneQuery(Query *query, int cursorOptions,
/* run it (if needed) and produce output */
ExplainOnePlan(plan, into, es, queryString, params, queryEnv,
- , (es->buffers ? : NULL));
+ , (es->buffers ? : NULL),
+ matviewInfo);
}
}
@@ -455,7 +454,7 @@ ExplainOneUtility(Nod
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sun, 07 Mar 2021 at 14:25, Bharath Rupireddy wrote: > On Sun, Mar 7, 2021 at 11:49 AM Japin Li wrote: >> >> On Fri, 05 Mar 2021 at 19:48, Bharath Rupireddy >> wrote: >> > Attaching v5 patch set for further review. >> > >> >> The v5 patch looks good to me, if there is no objection, I'll change the >> cf status to "Ready for Committer" in few days. > > Thanks for the review. > > As I mentioned upthread, I have 2 open points: > 1) In the patch I have added a new mat view info parameter to > ExplainOneQuery(), do we also need to add it to > ExplainOneQuery_hook_type? IMO, we should not (for now), because this > would create a backward compatibility issue. Sorry, I do not know how PostgreSQL handle the backward compatibility issue. Is there a guideline? > 2) Do we document (under respective command pages or somewhere else) > that we allow explain/explain analyze for a command? > IMO, we can add a new page to list the commands that can be explain/explain analyze, since it's clear for users. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sun, Mar 7, 2021 at 11:49 AM Japin Li wrote: > > On Fri, 05 Mar 2021 at 19:48, Bharath Rupireddy > wrote: > > Attaching v5 patch set for further review. > > > > The v5 patch looks good to me, if there is no objection, I'll change the > cf status to "Ready for Committer" in few days. Thanks for the review. As I mentioned upthread, I have 2 open points: 1) In the patch I have added a new mat view info parameter to ExplainOneQuery(), do we also need to add it to ExplainOneQuery_hook_type? IMO, we should not (for now), because this would create a backward compatibility issue. 2) Do we document (under respective command pages or somewhere else) that we allow explain/explain analyze for a command? Thoughts? With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Fri, 05 Mar 2021 at 19:48, Bharath Rupireddy wrote: > Attaching v5 patch set for further review. > The v5 patch looks good to me, if there is no objection, I'll change the cf status to "Ready for Committer" in few days. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
not a single action",
+ RelationGetRelationName(matviewRel));
+
+ /*
+ * Check that there is a unique index with no WHERE clause on one or more
+ * columns of the materialized view if CONCURRENTLY is specified.
+ */
+ if (stmt->concurrent)
+ {
+ List *indexoidlist = RelationGetIndexList(matviewRel);
+ ListCell *indexoidscan;
+ bool hasUniqueIndex = false;
+
+ foreach(indexoidscan, indexoidlist)
+ {
+ Oid indexoid = lfirst_oid(indexoidscan);
+ Relation indexRel;
+
+ indexRel = index_open(indexoid, AccessShareLock);
+ hasUniqueIndex = is_usable_unique_index(indexRel);
+ index_close(indexRel, AccessShareLock);
+ if (hasUniqueIndex)
+break;
+ }
+
+ list_free(indexoidlist);
+
+ if (!hasUniqueIndex)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("cannot refresh materialized view \"%s\" concurrently",
+ quote_qualified_identifier(get_namespace_name(RelationGetNamespace(matviewRel)),
+ RelationGetRelationName(matviewRel))),
+ errhint("Create a unique index with no WHERE clause on one or more columns of the materialized view.")));
+ }
+
+ /*
+ * The stored query was rewritten at the time of the MV definition, but
+ * has not been scribbled on by the planner.
+ */
+ dataQuery = linitial_node(Query, actions);
+
+ /*
+ * Check for active uses of the relation in the current transaction, such
+ * as open scans.
+ *
+ * NB: We count on this to protect us against problems with refreshing the
+ * data using TABLE_INSERT_FROZEN.
+ */
+ CheckTableNotInUse(matviewRel, "REFRESH MATERIALIZED VIEW");
+
+ *rel = matviewRel;
+ *objectId = matviewOid;
+
+ return dataQuery;
+}
+
+/*
+ * get_new_heap_oid
+ *
+ * Create a new heap and return its oid to which the refresh materialized view
+ * data is inserted into.
+ */
+static Oid
+get_new_heap_oid(RefreshMatViewStmt *stmt, Relation matviewRel, Oid matviewOid,
+ char *relpersistence)
+{
+ Oid OIDNewHeap;
+ Oid tableSpace;
+
+ /* Concurrent refresh builds new data in temp tablespace, and does diff. */
+ if (stmt->concurrent)
+ {
+ tableSpace = GetDefaultTablespace(RELPERSISTENCE_TEMP, false);
+ *relpersistence = RELPERSISTENCE_TEMP;
+ }
+ else
+ {
+ tableSpace = matviewRel->rd_rel->reltablespace;
+ *relpersistence = matviewRel->rd_rel->relpersistence;
+ }
+
+ /*
+ * Create the transient table that will receive the regenerated data. Lock
+ * it against access by any other process until commit (by which time it
+ * will be gone).
+ */
+ OIDNewHeap = make_new_heap(matviewOid, tableSpace, *relpersistence,
+ ExclusiveLock);
+ LockRelationOid(OIDNewHeap, AccessExclusiveLock);
+
+ return OIDNewHeap;
+}
+
+/*
+ * match_matview_with_new_data
+ *
+ * Arrange the materialized view newly generated data to match the existing
+ * data i.e merge in case of CONCURRENTLY otherwise perform heap swap and
+ * truncate the materialized view.
+ */
+static void
+match_matview_with_new_data(RefreshMatViewStmt *stmt, Relation matviewRel,
+ Oid matviewOid, Oid OIDNewHeap, Oid relowner,
+ int save_sec_context, char relpersistence,
+ uint64 processed)
+{
+ /* Make the materialized view match the newly generated data. */
+ if (stmt->concurrent)
+ {
+ int old_depth = matview_maintenance_depth;
+
+ PG_TRY();
+ {
+ refresh_by_match_merge(matviewOid, OIDNewHeap, relowner,
+ save_sec_context);
+ }
+ PG_CATCH();
+ {
+ matview_maintenance_depth = old_depth;
+ PG_RE_THROW();
+ }
+ PG_END_TRY();
+ Assert(matview_maintenance_depth == old_depth);
+ }
+ else
+ {
+ refresh_by_heap_swap(matviewOid, OIDNewHeap, relpersistence);
+
+ /*
+ * Inform stats collector about our activity: basically, we truncated
+ * the materialized view and inserted some new data. (The concurrent
+ * code path above doesn't need to worry about this because the
+ * inserts and deletes it issues get counted by lower-level code.)
+ */
+ pgstat_count_truncate(matviewRel);
+ if (!stmt->skipData)
+ pgstat_count_heap_insert(matviewRel, processed);
+ }
+}
+
+/*
+ * rewrite_refresh_matview_query
+ *
+ * Rewrite the refresh materialized view data generating query.
+ *
+ * Work on the copied query to preserve the original query. Because the
+ * rewriter and planner tend to scribble on the input, we make a preliminary
+ * copy of the source querytree. This prevents problems in the case that
+ * REFRESH MATERIALIZED VIEW is in a portal or plpgsql function and is executed
+ * repeatedly. (See also the same hack in EXPLAIN and PREPARE.)
+ */
+static Query *
+rewrite_refresh_matview_query(Query *dataQuery)
+{
+ List *rewritten;
+ Query *copied_query;
+
+ /* Lock and rewrite, using a copy to preserve the original query. */
+ copied_query = copyObject(dataQuery);
+ AcquireRewriteLocks(copied_query, true, false);
+ rewritten = QueryRewrite(copied_query);
+
+ /* SELECT should never rewr
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Thu, 04 Mar 2021 at 14:53, Bharath Rupireddy
wrote:
> Thanks! I will look forward for more review comments.
>
v4-0001-Rearrange-Refresh-Mat-View-Code.patch
-
+static Oid
+get_new_heap_oid(RefreshMatViewStmt *stmt, Relation matviewRel, Oid matviewOid,
+char *relpersistence)
+{
+ Oid OIDNewHeap;
+ boolconcurrent;
+ Oid tableSpace;
+
+ concurrent = stmt->concurrent;
+
+ /* Concurrent refresh builds new data in temp tablespace, and does
diff. */
+ if (concurrent)
+ {
+ tableSpace = GetDefaultTablespace(RELPERSISTENCE_TEMP, false);
+ *relpersistence = RELPERSISTENCE_TEMP;
+ }
Since the concurrent only use in one place, I think we can remove the local
variable
concurrent in get_new_heap_oid().
The others looks good to me.
--
Regrads,
Japin Li.
ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Thu, Mar 4, 2021 at 11:41 AM Japin Li wrote: > On Wed, 03 Mar 2021 at 20:56, David Steele wrote: > > Do you know if you will have time to review this patch during the > > current commitfest? > > > > Sorry for the late reply! I think I have time to review this patch > and I will do it later. Thanks! I will look forward for more review comments. With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Wed, 03 Mar 2021 at 20:56, David Steele wrote: > Do you know if you will have time to review this patch during the > current commitfest? > Sorry for the late reply! I think I have time to review this patch and I will do it later. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
Hi Japin, On 1/8/21 9:02 PM, japin wrote: On Sat, 09 Jan 2021 at 09:38, Bharath Rupireddy wrote: On Fri, Jan 8, 2021 at 9:50 PM japin wrote: I ran pgindent on 0001 patch to fix the above. 0002 patch has no changes. If I'm correct, pgindent will be run periodically on master. Thanks for your point out. I don't know before. Do you know if you will have time to review this patch during the current commitfest? Regards, -- -David da...@pgmasters.net
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Sat, 09 Jan 2021 at 09:38, Bharath Rupireddy wrote: > On Fri, Jan 8, 2021 at 9:50 PM japin wrote: >> > Attaching v3 patches, please consider these for further review. >> > >> >> I find that both the declaration and definition of >> match_matview_with_new_data() >> have a tab between type and variable. We can use pgindent to fix it. >> What do you think? >> >> >> static void >> match_matview_with_new_data(RefreshMatViewStmt *stmt, Relation matviewRel, >> ^ >> Oid matviewOid, Oid OIDNewHeap, Oid relowner, >> int save_sec_context, char relpersistence, >> uint64 processed) >> ^ > > I ran pgindent on 0001 patch to fix the above. 0002 patch has no > changes. If I'm correct, pgindent will be run periodically on master. > Thanks for your point out. I don't know before. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Fri, Jan 8, 2021 at 9:50 PM japin wrote: > > Attaching v3 patches, please consider these for further review. > > > > I find that both the declaration and definition of > match_matview_with_new_data() > have a tab between type and variable. We can use pgindent to fix it. > What do you think? > > > static void > match_matview_with_new_data(RefreshMatViewStmt *stmt, Relation matviewRel, > ^ > Oid matviewOid, Oid OIDNewHeap, Oid relowner, > int save_sec_context, char relpersistence, > uint64 processed) > ^ I ran pgindent on 0001 patch to fix the above. 0002 patch has no changes. If I'm correct, pgindent will be run periodically on master. Attaching v4 patch set for further review. With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com v4-0001-Rearrange-Refresh-Mat-View-Code.patch Description: Binary data v4-0002-EXPLAIN-EXPLAIN-ANALYZE-REFRESH-MATERIALIZED-VIEW.patch Description: Binary data
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Fri, 08 Jan 2021 at 17:24, Bharath Rupireddy wrote: > On Fri, Jan 8, 2021 at 1:50 PM japin wrote: >> Thanks for updating the patch! >> >> + /* Get the data generating query. */ >> + dataQuery = get_matview_query(stmt, , ); >> >> - /* >> -* Check for active uses of the relation in the current transaction, >> such >> -* as open scans. >> -* >> -* NB: We count on this to protect us against problems with >> refreshing the >> -* data using TABLE_INSERT_FROZEN. >> -*/ >> - CheckTableNotInUse(matviewRel, "REFRESH MATERIALIZED VIEW"); >> + relowner = matviewRel->rd_rel->relowner; >> >> After apply the patch, there is a duplicate >> >> relowner = matviewRel->rd_rel->relowner; > > Corrected that. > >> + else if(matviewInfo) >> + dest = >> CreateTransientRelDestReceiver(matviewInfo->OIDNewHeap); >> >> If the `matviewInfo->OIDNewHeap` is invalid, IMO we don't need create >> DestReceiver, isn't it? And we should add a space after `if`. > > Yes, we can skip creating the dest receiver when OIDNewHeap is > invalid, this can happen for plain explain refresh mat view case. > > if (explainInfo && !explainInfo->es->analyze) > OIDNewHeap = InvalidOid; > else > OIDNewHeap = get_new_heap_oid(stmt, matviewRel, matviewOid, > ); > > Since we don't call ExecutorRun for plain explain, we can skip the > dest receiver creation. I modified the code as below in explain.c. > > if (into) > dest = CreateIntoRelDestReceiver(into); > else if (matviewInfo && OidIsValid(matviewInfo->OIDNewHeap)) > dest = CreateTransientRelDestReceiver(matviewInfo->OIDNewHeap); > else > dest = None_Receiver; > > Thanks for taking a look at the patches. > Thanks! > Attaching v3 patches, please consider these for further review. > I find that both the declaration and definition of match_matview_with_new_data() have a tab between type and variable. We can use pgindent to fix it. What do you think? static void match_matview_with_new_data(RefreshMatViewStmt *stmt, Relation matviewRel, ^ Oid matviewOid, Oid OIDNewHeap, Oid relowner, int save_sec_context, char relpersistence, uint64 processed) ^ -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.,Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
query.
+ *
+ * Work on the copied query to preserve the original query. Because the
+ * rewriter and planner tend to scribble on the input, we make a preliminary
+ * copy of the source querytree. This prevents problems in the case that
+ * REFRESH MATERIALIZED VIEW is in a portal or plpgsql function and is executed
+ * repeatedly. (See also the same hack in EXPLAIN and PREPARE.)
+ */
+static Query *
+rewrite_refresh_matview_query(Query *dataQuery)
+{
+ List *rewritten;
+ Query *copied_query;
+
+ /* Lock and rewrite, using a copy to preserve the original query. */
+ copied_query = copyObject(dataQuery);
+ AcquireRewriteLocks(copied_query, true, false);
+ rewritten = QueryRewrite(copied_query);
+
+ /* SELECT should never rewrite to more or less than one SELECT query */
+ if (list_length(rewritten) != 1)
+ elog(ERROR, "unexpected rewrite result for REFRESH MATERIALIZED VIEW");
+ dataQuery = (Query *) linitial(rewritten);
+
+ return dataQuery;
+}
--
2.25.1
From c627d596cc4ac06e406952ed97b3e417b857aedd Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy
Date: Fri, 8 Jan 2021 14:15:46 +0530
Subject: [PATCH v3 2/2] EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
Currently, explain/explain analyze refresh materialized view(RMV)
is not allowed. We do plan the materialized view query before every
refresh. I propose to show the explain/explain analyze of the
select part of the materialized view. It will be useful for the
user to know what exactly is being planned and executed as part of
RMV. Please note that we already have explain/explain analyze
CTAS/Create Mat View(CMV), where we show the explain/explain analyze
of the select part. This proposal will do the same thing.
The behaviour can be like this:
EXPLAIN REFRESH MATERIALIZED VIEW mv1; --> will not refresh the mat
view, but shows the select part's plan of mat view.
EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW mv1; --> will refresh the
mat view and shows the select part's plan of mat view.
---
src/backend/commands/explain.c| 50 +++-
src/backend/commands/matview.c| 36 +---
src/backend/commands/prepare.c| 3 +-
src/backend/tcop/utility.c| 3 +-
src/include/commands/explain.h| 30 +-
src/include/commands/matview.h| 8 ++-
src/test/regress/expected/matview.out | 83 +++
src/test/regress/sql/matview.sql | 27 +
8 files changed, 215 insertions(+), 25 deletions(-)
diff --git a/src/backend/commands/explain.c b/src/backend/commands/explain.c
index 5d7eb3574c..7343a916b0 100644
--- a/src/backend/commands/explain.c
+++ b/src/backend/commands/explain.c
@@ -17,6 +17,7 @@
#include "catalog/pg_type.h"
#include "commands/createas.h"
#include "commands/defrem.h"
+#include "commands/matview.h"
#include "commands/prepare.h"
#include "executor/nodeHash.h"
#include "foreign/fdwapi.h"
@@ -53,10 +54,6 @@ explain_get_index_name_hook_type explain_get_index_name_hook = NULL;
#define X_CLOSE_IMMEDIATE 2
#define X_NOWHITESPACE 4
-static void ExplainOneQuery(Query *query, int cursorOptions,
- IntoClause *into, ExplainState *es,
- const char *queryString, ParamListInfo params,
- QueryEnvironment *queryEnv);
static void ExplainPrintJIT(ExplainState *es, int jit_flags,
JitInstrumentation *ji);
static void report_triggers(ResultRelInfo *rInfo, bool show_relname,
@@ -274,7 +271,8 @@ ExplainQuery(ParseState *pstate, ExplainStmt *stmt,
{
ExplainOneQuery(lfirst_node(Query, l),
CURSOR_OPT_PARALLEL_OK, NULL, es,
- pstate->p_sourcetext, params, pstate->p_queryEnv);
+ pstate->p_sourcetext, params, pstate->p_queryEnv,
+ NULL);
/* Separate plans with an appropriate separator */
if (lnext(rewritten, l) != NULL)
@@ -357,11 +355,11 @@ ExplainResultDesc(ExplainStmt *stmt)
*
* "into" is NULL unless we are explaining the contents of a CreateTableAsStmt.
*/
-static void
+void
ExplainOneQuery(Query *query, int cursorOptions,
IntoClause *into, ExplainState *es,
const char *queryString, ParamListInfo params,
-QueryEnvironment *queryEnv)
+QueryEnvironment *queryEnv, RefreshMatViewInfo *matviewInfo)
{
/* planner will not cope with utility statements */
if (query->commandType == CMD_UTILITY)
@@ -402,7 +400,8 @@ ExplainOneQuery(Query *query, int cursorOptions,
/* run it (if needed) and produce output */
ExplainOnePlan(plan, into, es, queryString, params, queryEnv,
- , (es->buffers ? : NULL));
+ , (es->buffers ? : NULL),
+ matviewInfo);
}
}
@@ -455,7 +454,7 @@ ExplainOneUtility(Node *utilityStmt, IntoClause *into, ExplainState *es,
Assert(list_length(rewritten) == 1);
ExplainOneQuery(linitial_node(Query, rewritten),
CURSOR_OPT_PARALLEL_OK, ctas->into, es,
- queryString, params, queryEnv);
+ queryString, par
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Thu, 07 Jan 2021 at 17:53, Bharath Rupireddy wrote: > On Mon, Dec 28, 2020 at 5:56 PM Bharath Rupireddy > wrote: >> >> On Tue, Dec 22, 2020 at 7:01 PM Bharath Rupireddy >> wrote: >> > Currently, $subject is not allowed. We do plan the mat view query >> > before every refresh. I propose to show the explain/explain analyze of >> > the select part of the mat view in case of Refresh Mat View(RMV). It >> > will be useful for the user to know what exactly is being planned and >> > executed as part of RMV. Please note that we already have >> > explain/explain analyze CTAS/Create Mat View(CMV), where we show the >> > explain/explain analyze of the select part. This proposal will do the >> > same thing. >> > >> > The behaviour can be like this: >> > EXPLAIN REFRESH MATERIALIZED VIEW mv1; --> will not refresh the mat >> > view, but shows the select part's plan of mat view. >> > EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW mv1; --> will refresh the >> > mat view and shows the select part's plan of mat view. >> > >> > Thoughts? If okay, I will post a patch later. >> >> Attaching below patches: >> >> 0001 - Rearrange Refresh Mat View Code - Currently, the function >> ExecRefreshMatView in matview.c is having many lines of code which is >> not at all good from readability and maintainability perspectives. >> This patch adds a few functions and moves the code from >> ExecRefreshMatView to them making the code look better. >> >> 0002 - EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW support and tests. >> >> If this proposal is useful, I have few open points - 1) In the patch I >> have added a new mat view info parameter to ExplainOneQuery(), do we >> also need to add it to ExplainOneQuery_hook_type? 2) Do we document >> (under respective command pages or somewhere else) that we allow >> explain/explain analyze for a command? >> >> Thoughts? > > Attaching v2 patch set reabsed on the latest master f7a1a805cb. And > also added an entry for upcoming commitfest - > https://commitfest.postgresql.org/32/2928/ > > Please consider the v2 patches for further review. > Thanks for updating the patch! + /* Get the data generating query. */ + dataQuery = get_matview_query(stmt, , ); - /* -* Check for active uses of the relation in the current transaction, such -* as open scans. -* -* NB: We count on this to protect us against problems with refreshing the -* data using TABLE_INSERT_FROZEN. -*/ - CheckTableNotInUse(matviewRel, "REFRESH MATERIALIZED VIEW"); + relowner = matviewRel->rd_rel->relowner; After apply the patch, there is a duplicate relowner = matviewRel->rd_rel->relowner; + else if(matviewInfo) + dest = CreateTransientRelDestReceiver(matviewInfo->OIDNewHeap); If the `matviewInfo->OIDNewHeap` is invalid, IMO we don't need create DestReceiver, isn't it? And we should add a space after `if`. -- Regrads, Japin Li. ChengDu WenWu Information Technology Co.Ltd.
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Mon, Dec 28, 2020 at 5:56 PM Bharath Rupireddy wrote: > > On Tue, Dec 22, 2020 at 7:01 PM Bharath Rupireddy > wrote: > > Currently, $subject is not allowed. We do plan the mat view query > > before every refresh. I propose to show the explain/explain analyze of > > the select part of the mat view in case of Refresh Mat View(RMV). It > > will be useful for the user to know what exactly is being planned and > > executed as part of RMV. Please note that we already have > > explain/explain analyze CTAS/Create Mat View(CMV), where we show the > > explain/explain analyze of the select part. This proposal will do the > > same thing. > > > > The behaviour can be like this: > > EXPLAIN REFRESH MATERIALIZED VIEW mv1; --> will not refresh the mat > > view, but shows the select part's plan of mat view. > > EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW mv1; --> will refresh the > > mat view and shows the select part's plan of mat view. > > > > Thoughts? If okay, I will post a patch later. > > Attaching below patches: > > 0001 - Rearrange Refresh Mat View Code - Currently, the function > ExecRefreshMatView in matview.c is having many lines of code which is > not at all good from readability and maintainability perspectives. > This patch adds a few functions and moves the code from > ExecRefreshMatView to them making the code look better. > > 0002 - EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW support and tests. > > If this proposal is useful, I have few open points - 1) In the patch I > have added a new mat view info parameter to ExplainOneQuery(), do we > also need to add it to ExplainOneQuery_hook_type? 2) Do we document > (under respective command pages or somewhere else) that we allow > explain/explain analyze for a command? > > Thoughts? Attaching v2 patch set reabsed on the latest master f7a1a805cb. And also added an entry for upcoming commitfest - https://commitfest.postgresql.org/32/2928/ Please consider the v2 patches for further review. With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com v2-0001-Rearrange-Refresh-Mat-View-Code.patch Description: Binary data v2-0002-EXPLAIN-EXPLAIN-ANALYZE-REFRESH-MATERIALIZED-VIEW.patch Description: Binary data
Re: EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
On Tue, Dec 22, 2020 at 7:01 PM Bharath Rupireddy
wrote:
> Currently, $subject is not allowed. We do plan the mat view query
> before every refresh. I propose to show the explain/explain analyze of
> the select part of the mat view in case of Refresh Mat View(RMV). It
> will be useful for the user to know what exactly is being planned and
> executed as part of RMV. Please note that we already have
> explain/explain analyze CTAS/Create Mat View(CMV), where we show the
> explain/explain analyze of the select part. This proposal will do the
> same thing.
>
> The behaviour can be like this:
> EXPLAIN REFRESH MATERIALIZED VIEW mv1; --> will not refresh the mat
> view, but shows the select part's plan of mat view.
> EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW mv1; --> will refresh the
> mat view and shows the select part's plan of mat view.
>
> Thoughts? If okay, I will post a patch later.
Attaching below patches:
0001 - Rearrange Refresh Mat View Code - Currently, the function
ExecRefreshMatView in matview.c is having many lines of code which is
not at all good from readability and maintainability perspectives.
This patch adds a few functions and moves the code from
ExecRefreshMatView to them making the code look better.
0002 - EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW support and tests.
If this proposal is useful, I have few open points - 1) In the patch I
have added a new mat view info parameter to ExplainOneQuery(), do we
also need to add it to ExplainOneQuery_hook_type? 2) Do we document
(under respective command pages or somewhere else) that we allow
explain/explain analyze for a command?
Thoughts?
With Regards,
Bharath Rupireddy.
EnterpriseDB: http://www.enterprisedb.com
From e0422b72acbaed27182ad3816cdc921a0f962fe2 Mon Sep 17 00:00:00 2001
From: Bharath Rupireddy
Date: Fri, 25 Dec 2020 14:03:17 +0530
Subject: [PATCH v1] Rearrange Refresh Mat View Code
Currently, the function ExecRefreshMatView in matview.c is having
many lines of code which is not at all good from readability and
maintainability perspectives. This patch adds few functions and
moves the code from ExecRefreshMatView to them making the code
look better.
---
src/backend/commands/matview.c | 452 -
1 file changed, 273 insertions(+), 179 deletions(-)
diff --git a/src/backend/commands/matview.c b/src/backend/commands/matview.c
index cfc63915f3..40cb436d16 100644
--- a/src/backend/commands/matview.c
+++ b/src/backend/commands/matview.c
@@ -64,7 +64,7 @@ static void transientrel_startup(DestReceiver *self, int operation, TupleDesc ty
static bool transientrel_receive(TupleTableSlot *slot, DestReceiver *self);
static void transientrel_shutdown(DestReceiver *self);
static void transientrel_destroy(DestReceiver *self);
-static uint64 refresh_matview_datafill(DestReceiver *dest, Query *query,
+static uint64 refresh_matview_datafill(Oid OIDNewHeap, Query *query,
const char *queryString);
static char *make_temptable_name_n(char *tempname, int n);
static void refresh_by_match_merge(Oid matviewOid, Oid tempOid, Oid relowner,
@@ -73,6 +73,16 @@ static void refresh_by_heap_swap(Oid matviewOid, Oid OIDNewHeap, char relpersist
static bool is_usable_unique_index(Relation indexRel);
static void OpenMatViewIncrementalMaintenance(void);
static void CloseMatViewIncrementalMaintenance(void);
+static Query *get_matview_query(RefreshMatViewStmt *stmt, Relation *rel,
+Oid *objectId);
+static Query *rewrite_refresh_matview_query(Query *dataQuery);
+static Oid get_new_heap_oid(RefreshMatViewStmt *stmt, Relation matviewRel,
+ Oid matviewOid, char *relpersistence);
+static void match_matview_with_new_data(RefreshMatViewStmt *stmt,
+ Relation matviewRel, Oid matviewOid,
+ Oid OIDNewHeap, Oid relowner,
+ int save_sec_context,
+ char relpersistence, uint64 processed);
/*
* SetMatViewPopulatedState
@@ -140,127 +150,20 @@ ExecRefreshMatView(RefreshMatViewStmt *stmt, const char *queryString,
{
Oid matviewOid;
Relation matviewRel;
- RewriteRule *rule;
- List *actions;
Query *dataQuery;
- Oid tableSpace;
- Oid relowner;
Oid OIDNewHeap;
- DestReceiver *dest;
uint64 processed = 0;
- bool concurrent;
- LOCKMODE lockmode;
+ Oid relowner;
char relpersistence;
Oid save_userid;
int save_sec_context;
int save_nestlevel;
ObjectAddress address;
- /* Determine strength of lock needed. */
- concurrent = stmt->concurrent;
- lockmode = concurrent ? ExclusiveLock : AccessExclusiveLock;
-
- /*
- * Get a lock until end of transaction.
- */
- matviewOid = RangeVarGetRelidExtended(stmt->relation,
- lockmode, 0,
- RangeVarCallbackOwnsTable, NULL);
- matviewRel = table_open(matviewOid, NoLock);
-
- /* Make sure it is a materialized view. */
- if (matviewRel->rd_rel->relkind != RELKIND_MATVIEW)
- ereport(ERROR,
-(errcode(ERRCODE_
EXPLAIN/EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW
Hi, Currently, $subject is not allowed. We do plan the mat view query before every refresh. I propose to show the explain/explain analyze of the select part of the mat view in case of Refresh Mat View(RMV). It will be useful for the user to know what exactly is being planned and executed as part of RMV. Please note that we already have explain/explain analyze CTAS/Create Mat View(CMV), where we show the explain/explain analyze of the select part. This proposal will do the same thing. The behaviour can be like this: EXPLAIN REFRESH MATERIALIZED VIEW mv1; --> will not refresh the mat view, but shows the select part's plan of mat view. EXPLAIN ANALYZE REFRESH MATERIALIZED VIEW mv1; --> will refresh the mat view and shows the select part's plan of mat view. Thoughts? If okay, I will post a patch later. With Regards, Bharath Rupireddy. EnterpriseDB: http://www.enterprisedb.com
