public inbox for [email protected]
help / color / mirror / Atom feedFrom: Alvaro Herrera <[email protected]>
To: Tom Lane <[email protected]>
Cc: Andres Freund <[email protected]>
Cc: [email protected]
Cc: Jaime Casanova <[email protected]>
Subject: Re: LogwrtResult contended spinlock
Date: Tue, 5 Apr 2022 17:47:04 +0200
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
Here's a v8, where per my previous comment I removed some code that I
believe is no longer necessary.
I've omitted the patch that renames LogwrtResult subvariables into
LogWriteResult/LogWriteFlush; I still think the end result is better
after that one, but it's a pretty trivial change that can be dealt with
separately afterwards.
--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/
"Saca el libro que tu religión considere como el indicado para encontrar la
oración que traiga paz a tu alma. Luego rebootea el computador
y ve si funciona" (Carlos Duclós)
Attachments:
[text/x-diff] v8-0001-Split-LogwrtResult-into-separate-variables.patch (12.2K, ../[email protected]/2-v8-0001-Split-LogwrtResult-into-separate-variables.patch)
download | inline diff:
From fdca5f32573342d950feefbdd4d64da6dd1cfb34 Mon Sep 17 00:00:00 2001
From: Alvaro Herrera <[email protected]>
Date: Tue, 22 Mar 2022 16:49:40 +0100
Subject: [PATCH v8] Split LogwrtResult into separate variables
Since the schedule of updating each portion is independent, it's not
very useful to have them both as a single struct. Before we used
atomics for the corresponding XLogCtl struct this made sense because we
could use struct assignment, but that's no longer the case.
Suggested by Andres Freund.
---
src/backend/access/transam/xlog.c | 99 ++++++++++++++-----------------
src/include/port/atomics.h | 29 +++++++++
src/tools/pgindent/typedefs.list | 1 +
3 files changed, 74 insertions(+), 55 deletions(-)
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index 17a56152f1..f1cb725bed 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -285,16 +285,13 @@ static bool doPageWrites;
*
* LogwrtRqst indicates a byte position that we need to write and/or fsync
* the log up to (all records before that point must be written or fsynced).
- * LogwrtResult indicates the byte positions we have already written/fsynced.
- * These structs are identical but are declared separately to indicate their
- * slightly different functions.
+ * LogWrtResult indicates the byte positions we have already written/fsynced.
+ * These structs are similar but are declared separately to indicate their
+ * slightly different functions; in addition, the latter is read and written
+ * using atomic operations.
*
- * To read XLogCtl->LogwrtResult, you must hold either info_lck or
- * WALWriteLock. To update it, you need to hold both locks. The point of
- * this arrangement is that the value can be examined by code that already
- * holds WALWriteLock without needing to grab info_lck as well. In addition
- * to the shared variable, each backend has a private copy of LogwrtResult,
- * which is updated when convenient.
+ * In addition to the shared variable, each backend has a private copy of
+ * LogwrtResult, each member of which is separately updated when convenient.
*
* The request bookkeeping is simpler: there is a shared XLogCtl->LogwrtRqst
* (protected by info_lck), but we don't need to cache any copies of it.
@@ -317,6 +314,12 @@ static bool doPageWrites;
*----------
*/
+typedef struct XLogwrtAtomic
+{
+ pg_atomic_uint64 Write; /* last byte + 1 written out */
+ pg_atomic_uint64 Flush; /* last byte + 1 flushed */
+} XLogwrtAtomic;
+
typedef struct XLogwrtRqst
{
XLogRecPtr Write; /* last byte + 1 to write out */
@@ -480,6 +483,7 @@ typedef struct XLogCtlData
{
XLogCtlInsert Insert;
+ XLogwrtAtomic LogwrtResult; /* uses atomics */
/* Protected by info_lck: */
XLogwrtRqst LogwrtRqst;
XLogRecPtr RedoRecPtr; /* a recent copy of Insert->RedoRecPtr */
@@ -497,12 +501,6 @@ typedef struct XLogCtlData
pg_time_t lastSegSwitchTime;
XLogRecPtr lastSegSwitchLSN;
- /*
- * Protected by info_lck and WALWriteLock (you must hold either lock to
- * read it, but both to update)
- */
- XLogwrtResult LogwrtResult;
-
/*
* Latest initialized page in the cache (last byte position + 1).
*
@@ -622,7 +620,7 @@ static ControlFileData *ControlFile = NULL;
static int UsableBytesInSegment;
/*
- * Private, possibly out-of-date copy of shared LogwrtResult.
+ * Private, possibly out-of-date copy of shared XLogCtl->LogwrtResult.
* See discussion above.
*/
static XLogwrtResult LogwrtResult = {0, 0};
@@ -932,8 +930,6 @@ XLogInsertRecord(XLogRecData *rdata,
/* advance global request to include new block(s) */
if (XLogCtl->LogwrtRqst.Write < EndPos)
XLogCtl->LogwrtRqst.Write = EndPos;
- /* update local result copy while I have the chance */
- LogwrtResult = XLogCtl->LogwrtResult;
SpinLockRelease(&XLogCtl->info_lck);
}
@@ -1811,6 +1807,7 @@ AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
* Now that we have the lock, check if someone initialized the page
* already.
*/
+ LogwrtResult.Write = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
while (upto >= XLogCtl->InitializedUpTo || opportunistic)
{
nextidx = XLogRecPtrToBufIdx(XLogCtl->InitializedUpTo);
@@ -1830,17 +1827,18 @@ AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
if (opportunistic)
break;
- /* Before waiting, get info_lck and update LogwrtResult */
+ /* Advance shared memory write request position */
SpinLockAcquire(&XLogCtl->info_lck);
if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
- LogwrtResult = XLogCtl->LogwrtResult;
SpinLockRelease(&XLogCtl->info_lck);
/*
- * Now that we have an up-to-date LogwrtResult value, see if we
- * still need to write it or if someone else already did.
+ * Before waiting, update LogwrtResult.Write and see if we still need
+ * to write it or if someone else already did.
*/
+ LogwrtResult.Write =
+ pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
if (LogwrtResult.Write < OldPageRqstPtr)
{
/*
@@ -1855,7 +1853,8 @@ AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
- LogwrtResult = XLogCtl->LogwrtResult;
+ LogwrtResult.Write =
+ pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
if (LogwrtResult.Write >= OldPageRqstPtr)
{
/* OK, someone wrote it already */
@@ -2101,7 +2100,7 @@ XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
/*
* Update local LogwrtResult (caller probably did this already, but...)
*/
- LogwrtResult = XLogCtl->LogwrtResult;
+ LogwrtResult.Write = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
/*
* Since successive pages in the xlog cache are consecutively allocated,
@@ -2316,6 +2315,10 @@ XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
Assert(npages == 0);
+ /* Publish current write result position */
+ pg_atomic_monotonic_advance_u64(&XLogCtl->LogwrtResult.Write,
+ LogwrtResult.Write);
+
/*
* If asked to flush, do so
*/
@@ -2352,22 +2355,9 @@ XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
LogwrtResult.Flush = LogwrtResult.Write;
}
- /*
- * Update shared-memory status
- *
- * We make sure that the shared 'request' values do not fall behind the
- * 'result' values. This is not absolutely essential, but it saves some
- * code in a couple of places.
- */
- {
- SpinLockAcquire(&XLogCtl->info_lck);
- XLogCtl->LogwrtResult = LogwrtResult;
- if (XLogCtl->LogwrtRqst.Write < LogwrtResult.Write)
- XLogCtl->LogwrtRqst.Write = LogwrtResult.Write;
- if (XLogCtl->LogwrtRqst.Flush < LogwrtResult.Flush)
- XLogCtl->LogwrtRqst.Flush = LogwrtResult.Flush;
- SpinLockRelease(&XLogCtl->info_lck);
- }
+ /* Publish current flush result position */
+ pg_atomic_monotonic_advance_u64(&XLogCtl->LogwrtResult.Flush,
+ LogwrtResult.Flush);
}
/*
@@ -2381,8 +2371,8 @@ XLogSetAsyncXactLSN(XLogRecPtr asyncXactLSN)
XLogRecPtr WriteRqstPtr = asyncXactLSN;
bool sleeping;
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
SpinLockAcquire(&XLogCtl->info_lck);
- LogwrtResult = XLogCtl->LogwrtResult;
sleeping = XLogCtl->WalWriterSleeping;
if (XLogCtl->asyncXactLSN < asyncXactLSN)
XLogCtl->asyncXactLSN = asyncXactLSN;
@@ -2587,10 +2577,10 @@ XLogFlush(XLogRecPtr record)
SpinLockAcquire(&XLogCtl->info_lck);
if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
- LogwrtResult = XLogCtl->LogwrtResult;
SpinLockRelease(&XLogCtl->info_lck);
/* done already? */
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
if (record <= LogwrtResult.Flush)
break;
@@ -2618,7 +2608,7 @@ XLogFlush(XLogRecPtr record)
}
/* Got the lock; recheck whether request is satisfied */
- LogwrtResult = XLogCtl->LogwrtResult;
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
if (record <= LogwrtResult.Flush)
{
LWLockRelease(WALWriteLock);
@@ -2742,7 +2732,6 @@ XLogBackgroundFlush(void)
/* read LogwrtResult and update local state */
SpinLockAcquire(&XLogCtl->info_lck);
- LogwrtResult = XLogCtl->LogwrtResult;
WriteRqst = XLogCtl->LogwrtRqst;
SpinLockRelease(&XLogCtl->info_lck);
@@ -2750,8 +2739,10 @@ XLogBackgroundFlush(void)
WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
/* if we have already flushed that far, consider async commit records */
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
if (WriteRqst.Write <= LogwrtResult.Flush)
{
+ pg_memory_barrier();
SpinLockAcquire(&XLogCtl->info_lck);
WriteRqst.Write = XLogCtl->asyncXactLSN;
SpinLockRelease(&XLogCtl->info_lck);
@@ -2767,6 +2758,7 @@ XLogBackgroundFlush(void)
{
if (openLogFile >= 0)
{
+ LogwrtResult.Write = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
if (!XLByteInPrevSeg(LogwrtResult.Write, openLogSegNo,
wal_segment_size))
{
@@ -2826,7 +2818,8 @@ XLogBackgroundFlush(void)
/* now wait for any in-progress insertions to finish and get write lock */
WaitXLogInsertionsToFinish(WriteRqst.Write);
LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
- LogwrtResult = XLogCtl->LogwrtResult;
+ LogwrtResult.Write = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
if (WriteRqst.Write > LogwrtResult.Write ||
WriteRqst.Flush > LogwrtResult.Flush)
{
@@ -2914,9 +2907,7 @@ XLogNeedsFlush(XLogRecPtr record)
return false;
/* read LogwrtResult and update local state */
- SpinLockAcquire(&XLogCtl->info_lck);
- LogwrtResult = XLogCtl->LogwrtResult;
- SpinLockRelease(&XLogCtl->info_lck);
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
/* check again */
if (record <= LogwrtResult.Flush)
@@ -5489,7 +5480,9 @@ StartupXLOG(void)
LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
- XLogCtl->LogwrtResult = LogwrtResult;
+ /* XXX OK to write without WALWriteLock? */
+ pg_atomic_write_u64(&XLogCtl->LogwrtResult.Write, EndOfLog);
+ pg_atomic_write_u64(&XLogCtl->LogwrtResult.Flush, EndOfLog);
XLogCtl->LogwrtRqst.Write = EndOfLog;
XLogCtl->LogwrtRqst.Flush = EndOfLog;
@@ -5925,9 +5918,7 @@ GetFlushRecPtr(TimeLineID *insertTLI)
{
Assert(XLogCtl->SharedRecoveryState == RECOVERY_STATE_DONE);
- SpinLockAcquire(&XLogCtl->info_lck);
- LogwrtResult = XLogCtl->LogwrtResult;
- SpinLockRelease(&XLogCtl->info_lck);
+ LogwrtResult.Flush = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Flush);
/*
* If we're writing and flushing WAL, the time line can't be changing, so
@@ -9095,9 +9086,7 @@ GetXLogInsertRecPtr(void)
XLogRecPtr
GetXLogWriteRecPtr(void)
{
- SpinLockAcquire(&XLogCtl->info_lck);
- LogwrtResult = XLogCtl->LogwrtResult;
- SpinLockRelease(&XLogCtl->info_lck);
+ LogwrtResult.Write = pg_atomic_read_u64(&XLogCtl->LogwrtResult.Write);
return LogwrtResult.Write;
}
diff --git a/src/include/port/atomics.h b/src/include/port/atomics.h
index 9550e04aaa..7abed1785a 100644
--- a/src/include/port/atomics.h
+++ b/src/include/port/atomics.h
@@ -519,6 +519,35 @@ pg_atomic_sub_fetch_u64(volatile pg_atomic_uint64 *ptr, int64 sub_)
return pg_atomic_sub_fetch_u64_impl(ptr, sub_);
}
+/*
+ * Monotonically advance the given variable using only atomic operations until
+ * it's at least the target value.
+ *
+ * Full barrier semantics (even when value is unchanged).
+ */
+static inline void
+pg_atomic_monotonic_advance_u64(volatile pg_atomic_uint64 *ptr, uint64 target_)
+{
+ uint64 currval;
+
+#ifndef PG_HAVE_ATOMIC_U64_SIMULATION
+ AssertPointerAlignment(ptr, 8);
+#endif
+
+ currval = pg_atomic_read_u64(ptr);
+ if (currval >= target_)
+ {
+ pg_memory_barrier();
+ return;
+ }
+
+ while (currval < target_)
+ {
+ if (pg_atomic_compare_exchange_u64(ptr, &currval, target_))
+ break;
+ }
+}
+
#undef INSIDE_ATOMICS_H
#endif /* ATOMICS_H */
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 4b12c575ab..74e1070433 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -2975,6 +2975,7 @@ XLogRecoveryCtlData
XLogRedoAction
XLogSegNo
XLogSource
+XLogwrtAtomic
XLogwrtResult
XLogwrtRqst
XPVIV
--
2.30.2
view thread (2+ messages)
reply
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: [email protected]
Cc: [email protected], [email protected], [email protected], [email protected], [email protected]
Subject: Re: LogwrtResult contended spinlock
In-Reply-To: <[email protected]>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox