Better recheck of dead upstream servers. Previously nginx used to mark backend again as live as soon as fail_timeout passes (10s by default) since last failure. On the other hand, detecting dead backend takes up to 60s (proxy_connect_timeout) in typical situation "backend is down and doesn't respond to any packets". This resulted in suboptimal behaviour in the above situation (up to 23% of requests were directed to dead backend with default settings). More detailed description of the problem may be found here (in Russian): http://mailman.nginx.org/pipermail/nginx-ru/2011-August/042172.html Fix is to only allow one request after fail_timeout passes, and mark backend as "live" only if this request succeeds. Note that with new code backend will not be marked "live" unless "check" request is completed, and this may take a while in some specific workloads (e.g. streaming). This is believed to be acceptable.

/*
 * Copyright (C) Igor Sysoev
 */


#ifndef _NGX_MD5_H_INCLUDED_
#define _NGX_MD5_H_INCLUDED_


#include <ngx_config.h>
#include <ngx_core.h>


#if (NGX_HAVE_MD5)

#if (NGX_HAVE_OPENSSL_MD5_H)
#include <openssl/md5.h>
#else
#include <md5.h>
#endif


typedef MD5_CTX  ngx_md5_t;


#if (NGX_OPENSSL_MD5)

#define ngx_md5_init    MD5_Init
#define ngx_md5_update  MD5_Update
#define ngx_md5_final   MD5_Final

#else

#define ngx_md5_init    MD5Init
#define ngx_md5_update  MD5Update
#define ngx_md5_final   MD5Final

#endif


#else /* !NGX_HAVE_MD5 */


typedef struct {
    uint64_t  bytes;
    uint32_t  a, b, c, d;
    u_char    buffer[64];
} ngx_md5_t;


void ngx_md5_init(ngx_md5_t *ctx);
void ngx_md5_update(ngx_md5_t *ctx, const void *data, size_t size);
void ngx_md5_final(u_char result[16], ngx_md5_t *ctx);


#endif

#endif /* _NGX_MD5_H_INCLUDED_ */