From 303f0c6342c7728059a23eaf7c0e72eb1c40777e Mon Sep 17 00:00:00 2001
From: Konstantin Osipov <kostja@tarantool.org>
Date: Thu, 13 Aug 2015 00:32:59 +0300
Subject: [PATCH] gh-916: (limit of statements in transaction) review fixes
- update comments
- fix the logic of partial write recovery (still needs a test)
---
src/box/wal.cc | 14 +++++++-------
src/fio.c | 13 ++++++++-----
2 files changed, 15 insertions(+), 12 deletions(-)
diff --git a/src/box/wal.cc b/src/box/wal.cc
index 79fb5985a3..444028ec67 100644
--- a/src/box/wal.cc
+++ b/src/box/wal.cc
@@ -327,22 +327,22 @@ wal_write_to_disk(struct recovery_state *r, struct wal_writer *writer,
/*
* This code tries to write queued requests (=transactions) using as
- * less as possible I/O syscalls and memory copies. For this reason
+ * few I/O syscalls and memory copies as possible. For this reason
* writev(2) and `struct iovec[]` are used (see `struct fio_batch`).
*
* For each request (=transaction) each request row (=statement) is
* added to iov `batch`. A row can contain up to XLOG_IOVMAX iovecs.
* A request can have an **unlimited** number of rows. Since OS has
- * hardcoded limit up to `sysconf(_SC_IOV_MAX)` iovecs (usually 1024),
- * a single batch can't fit huge transactions. Therefore, it is not
- * possible to "atomically" write an entire transaction using the
- * single writev(2) call.
+ * a hard coded limit up to `sysconf(_SC_IOV_MAX)` iovecs (usually
+ * 1024), a huge transaction may not fit into a single batch.
+ * Therefore, it is not possible to "atomically" write an entire
+ * transaction using a single writev(2) call.
*
* Request boundaries and batch boundaries are not connected at all
* in this code. Batches flushed to disk as soon as they are full.
* In order to guarantee that a transaction is either fully written
* to file or isn't written at all, ftruncate(2) is used to shrink
- * file to the last fuly written request. The absolute position
+ * the file to the last fully written request. The absolute position
* of request in xlog file is stored inside `struct wal_request`.
*/
@@ -400,7 +400,7 @@ wal_write_to_disk(struct recovery_state *r, struct wal_writer *writer,
/* Flush remaining data in batch (if any) */
if (fio_batch_size(batch) > 0) {
ssize_t nwr = wal_fio_batch_write(batch, fileno(wal->f));
- if (nwr >= 0) {
+ if (nwr > 0) {
/* Update cached file offset */
written_bytes += nwr;
}
diff --git a/src/fio.c b/src/fio.c
index 8561d625a4..594ca49f69 100644
--- a/src/fio.c
+++ b/src/fio.c
@@ -221,13 +221,16 @@ fio_batch_rotate(struct fio_batch *batch, size_t bytes_written)
struct iovec *iov = batch->iov;
struct iovec *iovend = iov + batch->iovcnt;
- for (; bytes_written >= iov->iov_len; ++iov)
+ for (; iov < iovend; ++iov) {
+ if (iov->iov_len > bytes_written) {
+ iov->iov_base = (char *) iov->iov_base + bytes_written;
+ iov->iov_len -= bytes_written;
+ break;
+ }
bytes_written -= iov->iov_len;
-
+ }
assert(iov < iovend); /* Partial write */
- iov->iov_base = (char *) iov->iov_base + bytes_written;
- iov->iov_len -= bytes_written;
- memmove(batch->iov, iov, iovend - iov);
+ memmove(batch->iov, iov, (iovend - iov) * sizeof(struct iovec));
batch->iovcnt = iovend - iov;
}
--
GitLab