Mercurial > hg > nginx-quic
view src/os/unix/ngx_freebsd_sendfile_chain.c @ 4667:d05ab8793a69 stable-1.2
Merge of r4622, r4623: balancing changes.
*) Upstream: smooth weighted round-robin balancing.
For edge case weights like { 5, 1, 1 } we now produce { a, a, b, a, c, a, a }
sequence instead of { c, b, a, a, a, a, a } produced previously.
Algorithm is as follows: on each peer selection we increase current_weight
of each eligible peer by its weight, select peer with greatest current_weight
and reduce its current_weight by total number of weight points distributed
among peers.
In case of { 5, 1, 1 } weights this gives the following sequence of
current_weight's:
a b c
0 0 0 (initial state)
5 1 1 (a selected)
-2 1 1
3 2 2 (a selected)
-4 2 2
1 3 3 (b selected)
1 -4 3
6 -3 4 (a selected)
-1 -3 4
4 -2 5 (c selected)
4 -2 -2
9 -1 -1 (a selected)
2 -1 -1
7 0 0 (a selected)
0 0 0
To preserve weight reduction in case of failures the effective_weight
variable was introduced, which usually matches peer's weight, but is
reduced temporarily on peer failures.
This change also fixes loop with backup servers and proxy_next_upstream
http_404 (ticket #47), and skipping alive upstreams in some cases if there
are multiple dead ones (ticket #64).
*) Upstream: fixed ip_hash rebalancing with the "down" flag.
Due to weight being set to 0 for down peers, order of peers after sorting
wasn't the same as without the "down" flag (with down peers at the end),
resulting in client rebalancing for clients on other servers. The only
rebalancing which should happen after adding "down" to a server is one
for clients on the server.
The problem was introduced in r1377 (which fixed endless loop by setting
weight to 0 for down servers). The loop is no longer possible with new
smooth algorithm, so preserving original weight is safe.
author | Maxim Dounin <mdounin@mdounin.ru> |
---|---|
date | Mon, 04 Jun 2012 11:21:58 +0000 |
parents | 089cc5154c1e |
children | ad137a80919f |
line wrap: on
line source
/* * Copyright (C) Igor Sysoev * Copyright (C) Nginx, Inc. */ #include <ngx_config.h> #include <ngx_core.h> #include <ngx_event.h> /* * Although FreeBSD sendfile() allows to pass a header and a trailer, * it cannot send a header with a part of the file in one packet until * FreeBSD 5.3. Besides, over the fast ethernet connection sendfile() * may send the partially filled packets, i.e. the 8 file pages may be sent * as the 11 full 1460-bytes packets, then one incomplete 324-bytes packet, * and then again the 11 full 1460-bytes packets. * * Therefore we use the TCP_NOPUSH option (similar to Linux's TCP_CORK) * to postpone the sending - it not only sends a header and the first part of * the file in one packet, but also sends the file pages in the full packets. * * But until FreeBSD 4.5 turning TCP_NOPUSH off does not flush a pending * data that less than MSS, so that data may be sent with 5 second delay. * So we do not use TCP_NOPUSH on FreeBSD prior to 4.5, although it can be used * for non-keepalive HTTP connections. */ #if (IOV_MAX > 64) #define NGX_HEADERS 64 #define NGX_TRAILERS 64 #else #define NGX_HEADERS IOV_MAX #define NGX_TRAILERS IOV_MAX #endif ngx_chain_t * ngx_freebsd_sendfile_chain(ngx_connection_t *c, ngx_chain_t *in, off_t limit) { int rc, flags; u_char *prev; off_t size, send, prev_send, aligned, sent, fprev; size_t header_size, file_size; ngx_uint_t eintr, eagain, complete; ngx_err_t err; ngx_buf_t *file; ngx_array_t header, trailer; ngx_event_t *wev; ngx_chain_t *cl; struct sf_hdtr hdtr; struct iovec *iov, headers[NGX_HEADERS], trailers[NGX_TRAILERS]; wev = c->write; if (!wev->ready) { return in; } #if (NGX_HAVE_KQUEUE) if ((ngx_event_flags & NGX_USE_KQUEUE_EVENT) && wev->pending_eof) { (void) ngx_connection_error(c, wev->kq_errno, "kevent() reported about an closed connection"); wev->error = 1; return NGX_CHAIN_ERROR; } #endif /* the maximum limit size is the maximum size_t value - the page size */ if (limit == 0 || limit > (off_t) (NGX_MAX_SIZE_T_VALUE - ngx_pagesize)) { limit = NGX_MAX_SIZE_T_VALUE - ngx_pagesize; } send = 0; eagain = 0; flags = 0; header.elts = headers; header.size = sizeof(struct iovec); header.nalloc = NGX_HEADERS; header.pool = c->pool; trailer.elts = trailers; trailer.size = sizeof(struct iovec); trailer.nalloc = NGX_TRAILERS; trailer.pool = c->pool; for ( ;; ) { file = NULL; file_size = 0; header_size = 0; eintr = 0; complete = 0; prev_send = send; header.nelts = 0; trailer.nelts = 0; /* create the header iovec and coalesce the neighbouring bufs */ prev = NULL; iov = NULL; for (cl = in; cl && send < limit; cl = cl->next) { if (ngx_buf_special(cl->buf)) { continue; } if (!ngx_buf_in_memory_only(cl->buf)) { break; } size = cl->buf->last - cl->buf->pos; if (send + size > limit) { size = limit - send; } if (prev == cl->buf->pos) { iov->iov_len += (size_t) size; } else { if (header.nelts >= IOV_MAX){ break; } iov = ngx_array_push(&header); if (iov == NULL) { return NGX_CHAIN_ERROR; } iov->iov_base = (void *) cl->buf->pos; iov->iov_len = (size_t) size; } prev = cl->buf->pos + (size_t) size; header_size += (size_t) size; send += size; } if (cl && cl->buf->in_file && send < limit) { file = cl->buf; /* coalesce the neighbouring file bufs */ do { size = cl->buf->file_last - cl->buf->file_pos; if (send + size > limit) { size = limit - send; aligned = (cl->buf->file_pos + size + ngx_pagesize - 1) & ~((off_t) ngx_pagesize - 1); if (aligned <= cl->buf->file_last) { size = aligned - cl->buf->file_pos; } } file_size += (size_t) size; send += size; fprev = cl->buf->file_pos + size; cl = cl->next; } while (cl && cl->buf->in_file && send < limit && file->file->fd == cl->buf->file->fd && fprev == cl->buf->file_pos); } if (file) { /* create the trailer iovec and coalesce the neighbouring bufs */ prev = NULL; iov = NULL; while (cl && send < limit) { if (ngx_buf_special(cl->buf)) { cl = cl->next; continue; } if (!ngx_buf_in_memory_only(cl->buf)) { break; } size = cl->buf->last - cl->buf->pos; if (send + size > limit) { size = limit - send; } if (prev == cl->buf->pos) { iov->iov_len += (size_t) size; } else { if (trailer.nelts >= IOV_MAX){ break; } iov = ngx_array_push(&trailer); if (iov == NULL) { return NGX_CHAIN_ERROR; } iov->iov_base = (void *) cl->buf->pos; iov->iov_len = (size_t) size; } prev = cl->buf->pos + (size_t) size; send += size; cl = cl->next; } } if (file) { if (ngx_freebsd_use_tcp_nopush && c->tcp_nopush == NGX_TCP_NOPUSH_UNSET) { if (ngx_tcp_nopush(c->fd) == NGX_ERROR) { err = ngx_errno; /* * there is a tiny chance to be interrupted, however, * we continue a processing without the TCP_NOPUSH */ if (err != NGX_EINTR) { wev->error = 1; (void) ngx_connection_error(c, err, ngx_tcp_nopush_n " failed"); return NGX_CHAIN_ERROR; } } else { c->tcp_nopush = NGX_TCP_NOPUSH_SET; ngx_log_debug0(NGX_LOG_DEBUG_EVENT, c->log, 0, "tcp_nopush"); } } /* * sendfile() does unneeded work if sf_hdtr's count is 0, * but corresponding pointer is not NULL */ hdtr.headers = header.nelts ? (struct iovec *) header.elts: NULL; hdtr.hdr_cnt = header.nelts; hdtr.trailers = trailer.nelts ? (struct iovec *) trailer.elts: NULL; hdtr.trl_cnt = trailer.nelts; /* * the "nbytes bug" of the old sendfile() syscall: * http://www.freebsd.org/cgi/query-pr.cgi?pr=33771 */ if (!ngx_freebsd_sendfile_nbytes_bug) { header_size = 0; } sent = 0; #if (NGX_HAVE_AIO_SENDFILE) flags = c->aio_sendfile ? SF_NODISKIO : 0; #endif rc = sendfile(file->file->fd, c->fd, file->file_pos, file_size + header_size, &hdtr, &sent, flags); if (rc == -1) { err = ngx_errno; switch (err) { case NGX_EAGAIN: eagain = 1; break; case NGX_EINTR: eintr = 1; break; #if (NGX_HAVE_AIO_SENDFILE) case NGX_EBUSY: c->busy_sendfile = file; break; #endif default: wev->error = 1; (void) ngx_connection_error(c, err, "sendfile() failed"); return NGX_CHAIN_ERROR; } ngx_log_debug1(NGX_LOG_DEBUG_EVENT, c->log, err, "sendfile() sent only %O bytes", sent); /* * sendfile() in FreeBSD 3.x-4.x may return value >= 0 * on success, although only 0 is documented */ } else if (rc >= 0 && sent == 0) { /* * if rc is OK and sent equal to zero, then someone * has truncated the file, so the offset became beyond * the end of the file */ ngx_log_error(NGX_LOG_ALERT, c->log, 0, "sendfile() reported that \"%s\" was truncated at %O", file->file->name.data, file->file_pos); return NGX_CHAIN_ERROR; } ngx_log_debug4(NGX_LOG_DEBUG_EVENT, c->log, 0, "sendfile: %d, @%O %O:%uz", rc, file->file_pos, sent, file_size + header_size); } else { rc = writev(c->fd, header.elts, header.nelts); ngx_log_debug2(NGX_LOG_DEBUG_EVENT, c->log, 0, "writev: %d of %uz", rc, header_size); if (rc == -1) { err = ngx_errno; switch (err) { case NGX_EAGAIN: break; case NGX_EINTR: eintr = 1; break; default: wev->error = 1; ngx_connection_error(c, err, "writev() failed"); return NGX_CHAIN_ERROR; } ngx_log_debug0(NGX_LOG_DEBUG_EVENT, c->log, err, "writev() not ready"); } sent = rc > 0 ? rc : 0; } if (send - prev_send == sent) { complete = 1; } c->sent += sent; for (cl = in; cl; cl = cl->next) { if (ngx_buf_special(cl->buf)) { continue; } if (sent == 0) { break; } size = ngx_buf_size(cl->buf); if (sent >= size) { sent -= size; if (ngx_buf_in_memory(cl->buf)) { cl->buf->pos = cl->buf->last; } if (cl->buf->in_file) { cl->buf->file_pos = cl->buf->file_last; } continue; } if (ngx_buf_in_memory(cl->buf)) { cl->buf->pos += (size_t) sent; } if (cl->buf->in_file) { cl->buf->file_pos += sent; } break; } #if (NGX_HAVE_AIO_SENDFILE) if (c->busy_sendfile) { return cl; } #endif if (eagain) { /* * sendfile() may return EAGAIN, even if it has sent a whole file * part, it indicates that the successive sendfile() call would * return EAGAIN right away and would not send anything. * We use it as a hint. */ wev->ready = 0; return cl; } if (eintr) { continue; } if (!complete) { wev->ready = 0; return cl; } if (send >= limit || cl == NULL) { return cl; } in = cl; } }