From a8ec9732ff892dff8146a1d0e637dd30de2dcf53 Mon Sep 17 00:00:00 2001 From: Melanie Plageman Date: Fri, 27 Feb 2026 16:33:40 -0500 Subject: [PATCH v43 09/10] 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. Setting the visibility map on-access can avoid write amplification caused by vacuum later needing to set the page all-visible, trigger a write and potentially FPI. It also allows more frequent index-only scans, since they require pages to be marked all-visible in the VM. 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 | 16 +++++++-- 5 files changed, 54 insertions(+), 19 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 66726b22de6..651efa0127a 100644 --- a/src/backend/access/heap/heapam_handler.c +++ b/src/backend/access/heap/heapam_handler.c @@ -148,7 +148,8 @@ 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->xs_base.flags & SO_HINT_REL_READ_ONLY)); } /* Obtain share-lock on the buffer so we can examine visibility */ @@ -2545,7 +2546,8 @@ 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 41bfb6711c1..235d21c1a41 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; @@ -232,7 +234,8 @@ 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); /* @@ -253,7 +256,8 @@ 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; @@ -336,6 +340,8 @@ heap_page_prune_opt(Relation relation, Buffer buffer, Buffer *vmbuffer) * current implementation. */ params.options = HEAP_PAGE_PRUNE_ALLOW_FAST_PATH; + if (rel_read_only) + params.options |= HEAP_PAGE_PRUNE_SET_VM; heap_page_prune_and_freeze(¶ms, &presult, &dummy_off_loc, NULL, NULL); @@ -392,6 +398,7 @@ 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); @@ -461,9 +468,8 @@ 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. @@ -919,21 +925,37 @@ heap_page_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 - * set the VM for now. - */ - 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) @@ -1165,7 +1187,7 @@ 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 f698c2d899b..24001b27387 100644 --- a/src/backend/access/heap/vacuumlazy.c +++ b/src/backend/access/heap/vacuumlazy.c @@ -2021,7 +2021,7 @@ 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 c6aec63a505..90ca5a2cfa8 100644 --- a/src/include/access/heapam.h +++ b/src/include/access/heapam.h @@ -43,6 +43,7 @@ #define HEAP_PAGE_PRUNE_MARK_UNUSED_NOW (1 << 0) #define HEAP_PAGE_PRUNE_FREEZE (1 << 1) #define HEAP_PAGE_PRUNE_ALLOW_FAST_PATH (1 << 2) +#define HEAP_PAGE_PRUNE_SET_VM (1 << 3) typedef struct BulkInsertStateData *BulkInsertState; typedef struct GlobalVisState GlobalVisState; @@ -95,7 +96,12 @@ typedef struct HeapScanDescData */ ParallelBlockTableScanWorkerData *rs_parallelworkerdata; - /* Current heap block's corresponding page in the visibility map */ + /* + * For sequential scans, bitmap heap scans, TID range scans, and sample + * scans. The current heap block's corresponding page in the visibility + * map. If the relation is not modified by the query, on-access pruning + * may set the VM. + */ Buffer rs_vmbuffer; /* these fields only used in page-at-a-time mode and for bitmap scans */ @@ -126,7 +132,11 @@ 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; } IndexFetchHeapData; @@ -431,7 +441,7 @@ 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, -- 2.43.0