From aa05e68336207dbb64c0468ab3f017f8f66f9e05 Mon Sep 17 00:00:00 2001 From: Melanie Plageman Date: Fri, 27 Feb 2026 16:33:40 -0500 Subject: [PATCH v38 11/12] Allow on-access pruning to set pages all-visible Many queries do not modify the underlying relation. For such queries, if on-access pruning occurs during the scan, we can check whether the page has become all-visible and update the visibility map accordingly. Previously, only vacuum and COPY FREEZE marked pages as all-visible or all-frozen. This commit implements on-access VM setting for sequential scans as well as for the underlying heap relation in index scans and bitmap heap scans. Author: Melanie Plageman Reviewed-by: Andres Freund Reviewed-by: Kirill Reshke Reviewed-by: Chao Li Discussion: https://postgr.es/m/flat/CAAKRu_ZMw6Npd_qm2KM%2BFwQ3cMOMx1Dh3VMhp8-V7SOLxdK9-g%40mail.gmail.com --- src/backend/access/heap/heapam.c | 3 +- src/backend/access/heap/heapam_handler.c | 6 ++- src/backend/access/heap/pruneheap.c | 46 ++++++++++++++----- src/backend/access/heap/vacuumlazy.c | 2 +- src/include/access/heapam.h | 12 +++-- .../t/035_standby_logical_decoding.pl | 3 +- 6 files changed, 52 insertions(+), 20 deletions(-) diff --git a/src/backend/access/heap/heapam.c b/src/backend/access/heap/heapam.c index 044f385e477..dbdf6521c42 100644 --- a/src/backend/access/heap/heapam.c +++ b/src/backend/access/heap/heapam.c @@ -633,7 +633,8 @@ heap_prepare_pagescan(TableScanDesc sscan) /* * Prune and repair fragmentation for the whole page, if possible. */ - heap_page_prune_opt(scan->rs_base.rs_rd, buffer, &scan->rs_vmbuffer); + heap_page_prune_opt(scan->rs_base.rs_rd, buffer, &scan->rs_vmbuffer, + (sscan->rs_flags & SO_HINT_REL_READ_ONLY)); /* * We must hold share lock on the buffer content while examining tuple diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c index d35b688d751..a083b69ffcd 100644 --- a/src/backend/access/heap/heapam_handler.c +++ b/src/backend/access/heap/heapam_handler.c @@ -147,7 +147,8 @@ heapam_index_fetch_tuple(struct IndexFetchTableData *scan, */ if (prev_buf != hscan->xs_cbuf) heap_page_prune_opt(hscan->xs_base.rel, hscan->xs_cbuf, - &hscan->xs_vmbuffer); + &hscan->xs_vmbuffer, + !hscan->modifies_base_rel); } /* Obtain share-lock on the buffer so we can examine visibility */ @@ -2542,7 +2543,8 @@ BitmapHeapScanNextBlock(TableScanDesc scan, /* * Prune and repair fragmentation for the whole page, if possible. */ - heap_page_prune_opt(scan->rs_rd, buffer, &hscan->rs_vmbuffer); + heap_page_prune_opt(scan->rs_rd, buffer, &hscan->rs_vmbuffer, + scan->rs_flags & SO_HINT_REL_READ_ONLY); /* * We must hold share lock on the buffer content while examining tuple diff --git a/src/backend/access/heap/pruneheap.c b/src/backend/access/heap/pruneheap.c index 9b5a0726f2b..3cdc1a36441 100644 --- a/src/backend/access/heap/pruneheap.c +++ b/src/backend/access/heap/pruneheap.c @@ -44,6 +44,8 @@ typedef struct bool mark_unused_now; /* whether to attempt freezing tuples */ bool attempt_freeze; + /* whether to attempt setting the VM */ + bool attempt_set_vm; struct VacuumCutoffs *cutoffs; Relation relation; @@ -213,7 +215,8 @@ static void page_verify_redirects(Page page); static bool heap_page_will_freeze(bool did_tuple_hint_fpi, bool do_prune, bool do_hint_prune, PruneState *prstate); -static bool heap_page_will_set_vm(PruneState *prstate, PruneReason reason); +static bool heap_page_will_set_vm(PruneState *prstate, PruneReason reason, + bool do_prune, bool do_freeze); /* * Optionally prune and repair fragmentation in the specified page. @@ -233,7 +236,8 @@ static bool heap_page_will_set_vm(PruneState *prstate, PruneReason reason); * unpinning *vmbuffer. */ void -heap_page_prune_opt(Relation relation, Buffer buffer, Buffer *vmbuffer) +heap_page_prune_opt(Relation relation, Buffer buffer, Buffer *vmbuffer, + bool rel_read_only) { Page page = BufferGetPage(buffer); TransactionId prune_xid; @@ -315,6 +319,8 @@ heap_page_prune_opt(Relation relation, Buffer buffer, Buffer *vmbuffer) * current implementation. */ params.options = 0; + if (rel_read_only) + params.options = HEAP_PAGE_PRUNE_SET_VM; heap_page_prune_and_freeze(¶ms, &presult, &dummy_off_loc, NULL, NULL); @@ -371,6 +377,7 @@ prune_freeze_setup(PruneFreezeParams *params, /* cutoffs must be provided if we will attempt freezing */ Assert(!(params->options & HEAP_PAGE_PRUNE_FREEZE) || params->cutoffs); prstate->attempt_freeze = (params->options & HEAP_PAGE_PRUNE_FREEZE) != 0; + prstate->attempt_set_vm = (params->options & HEAP_PAGE_PRUNE_SET_VM) != 0; prstate->cutoffs = params->cutoffs; prstate->relation = params->relation; prstate->block = BufferGetBlockNumber(params->buffer); @@ -440,9 +447,8 @@ prune_freeze_setup(PruneFreezeParams *params, * We track whether the page will be all-visible/all-frozen at the end of * pruning and freezing. While examining tuple visibility, we'll set * set_all_visible to false if there are tuples on the page not visible to - * all running and future transactions. set_all_visible is always - * maintained but only VACUUM will set the VM if the page ends up being - * all-visible. + * all running and future transactions. If enabled for this scan, we will + * set the VM if the page ends up being all-visible. * * We also keep track of the newest live XID, which is used to calculate * the snapshot conflict horizon for a WAL record setting the VM. @@ -873,21 +879,37 @@ heap_fix_vm_corruption(PruneState *prstate, OffsetNumber offnum) * This function does not actually set the VM bits or page-level visibility * hint, PD_ALL_VISIBLE. * + * This should be called only after do_freeze has been decided (and do_prune + * has been set), as these factor into our heuristic-based decision. + * * Returns true if one or both VM bits should be set and false otherwise. */ static bool -heap_page_will_set_vm(PruneState *prstate, PruneReason reason) +heap_page_will_set_vm(PruneState *prstate, PruneReason reason, + bool do_prune, bool do_freeze) { - /* - * Though on-access pruning maintains prstate->set_all_visible, we don't - * consider setting the VM. - */ - if (reason == PRUNE_ON_ACCESS) + if (!prstate->attempt_set_vm) return false; if (!prstate->set_all_visible) return false; + /* + * If this is an on-access call and we're not actually pruning, avoid + * setting the visibility map if it would newly dirty the heap page or, if + * the page is already dirty, if doing so would require including a + * full-page image (FPI) of the heap page in the WAL. This situation + * should be rare, as on-access pruning is only attempted when + * pd_prune_xid is valid. + */ + if (reason == PRUNE_ON_ACCESS && !do_prune && !do_freeze && + (!BufferIsDirty(prstate->buffer) || XLogCheckBufferNeedsBackup(prstate->buffer))) + { + prstate->set_all_visible = false; + prstate->set_all_frozen = false; + return false; + } + prstate->new_vmbits = VISIBILITYMAP_ALL_VISIBLE; if (prstate->set_all_frozen) @@ -1103,7 +1125,7 @@ heap_page_prune_and_freeze(PruneFreezeParams *params, Assert(!prstate.set_all_frozen || prstate.set_all_visible); Assert(!prstate.set_all_visible || (prstate.lpdead_items == 0)); - do_set_vm = heap_page_will_set_vm(&prstate, params->reason); + do_set_vm = heap_page_will_set_vm(&prstate, params->reason, do_prune, do_freeze); /* * new_vmbits should be 0 regardless of whether or not the page is diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c index 8bd178ae7e6..d2cae77b52a 100644 --- a/src/backend/access/heap/vacuumlazy.c +++ b/src/backend/access/heap/vacuumlazy.c @@ -2008,7 +2008,7 @@ lazy_scan_prune(LVRelState *vacrel, .buffer = buf, .vmbuffer = vmbuffer, .reason = PRUNE_VACUUM_SCAN, - .options = HEAP_PAGE_PRUNE_FREEZE, + .options = HEAP_PAGE_PRUNE_FREEZE | HEAP_PAGE_PRUNE_SET_VM, .vistest = vacrel->vistest, .cutoffs = &vacrel->cutoffs, }; diff --git a/src/include/access/heapam.h b/src/include/access/heapam.h index 978ea90ffa2..768d442c39c 100644 --- a/src/include/access/heapam.h +++ b/src/include/access/heapam.h @@ -42,6 +42,7 @@ /* "options" flag bits for heap_page_prune_and_freeze */ #define HEAP_PAGE_PRUNE_MARK_UNUSED_NOW (1 << 0) #define HEAP_PAGE_PRUNE_FREEZE (1 << 1) +#define HEAP_PAGE_PRUNE_SET_VM (1 << 2) typedef struct BulkInsertStateData *BulkInsertState; typedef struct GlobalVisState GlobalVisState; @@ -97,7 +98,8 @@ typedef struct HeapScanDescData /* * For sequential scans, bitmap heap scans, TID range scans, and sample * scans. The current heap block's corresponding page in the visibility - * map. + * map. If the relation is not modified by the query, on-access pruning + * may set the VM. */ Buffer rs_vmbuffer; @@ -129,7 +131,11 @@ typedef struct IndexFetchHeapData */ Buffer xs_cbuf; - /* Current heap block's corresponding page in the visibility map */ + /* + * Current heap block's corresponding page in the visibility map. For + * index scans that do not modify the underlying heap table, on-access + * pruning may set the VM on-access. + */ Buffer xs_vmbuffer; /* @@ -440,7 +446,7 @@ extern TransactionId heap_index_delete_tuples(Relation rel, /* in heap/pruneheap.c */ extern void heap_page_prune_opt(Relation relation, Buffer buffer, - Buffer *vmbuffer); + Buffer *vmbuffer, bool rel_read_only); extern void heap_page_prune_and_freeze(PruneFreezeParams *params, PruneFreezeResult *presult, OffsetNumber *off_loc, diff --git a/src/test/recovery/t/035_standby_logical_decoding.pl b/src/test/recovery/t/035_standby_logical_decoding.pl index d264a698ff6..a5536ba4ff6 100644 --- a/src/test/recovery/t/035_standby_logical_decoding.pl +++ b/src/test/recovery/t/035_standby_logical_decoding.pl @@ -296,6 +296,7 @@ wal_level = 'logical' max_replication_slots = 4 max_wal_senders = 4 autovacuum = off +hot_standby_feedback = on }); $node_primary->dump_info; $node_primary->start; @@ -748,7 +749,7 @@ check_pg_recvlogical_stderr($handle, $logstart = -s $node_standby->logfile; reactive_slots_change_hfs_and_wait_for_xmins('shared_row_removal_', - 'no_conflict_', 0, 1); + 'no_conflict_', 1, 0); # This should not trigger a conflict wait_until_vacuum_can_remove( -- 2.43.0