Compare commits

..
6 changed files with 181 additions and 208 deletions
+1 -1
View File
@@ -206,7 +206,7 @@ endif
REDIS_SERVER_NAME=redis-server
REDIS_SENTINEL_NAME=redis-sentinel
REDIS_SERVER_OBJ=adlist.o quicklist.o ae.o anet.o dict.o server.o sds.o zmalloc.o lzf_c.o lzf_d.o pqsort.o zipmap.o sha1.o ziplist.o release.o networking.o util.o object.o db.o replication.o rdb.o t_string.o t_list.o t_set.o t_zset.o t_hash.o config.o aof.o pubsub.o multi.o debug.o sort.o intset.o syncio.o cluster.o crc16.o endianconv.o slowlog.o scripting.o bio.o rio.o rand.o memtest.o crc64.o bitops.o sentinel.o notify.o setproctitle.o blocked.o hyperloglog.o latency.o sparkline.o redis-check-rdb.o redis-check-aof.o geo.o lazyfree.o module.o evict.o expire.o geohash.o geohash_helper.o childinfo.o defrag.o siphash.o rax.o t_stream.o listpack.o localtime.o lolwut.o lolwut5.o lolwut6.o acl.o gopher.o tracking.o connection.o tls.o sha256.o timeout.o
REDIS_SERVER_OBJ=adlist.o quicklist.o ae.o anet.o dict.o server.o sds.o zmalloc.o lzf_c.o lzf_d.o pqsort.o zipmap.o sha1.o ziplist.o release.o networking.o util.o object.o db.o replication.o rdb.o t_string.o t_list.o t_set.o t_zset.o t_hash.o config.o aof.o pubsub.o multi.o debug.o sort.o intset.o syncio.o cluster.o crc16.o endianconv.o slowlog.o scripting.o bio.o rio.o rand.o memtest.o crc64.o bitops.o sentinel.o notify.o setproctitle.o blocked.o hyperloglog.o latency.o sparkline.o redis-check-rdb.o redis-check-aof.o geo.o lazyfree.o module.o evict.o expire.o geohash.o geohash_helper.o childinfo.o defrag.o siphash.o rax.o t_stream.o listpack.o localtime.o lolwut.o lolwut5.o lolwut6.o acl.o gopher.o tracking.o connection.o tls.o sha256.o
REDIS_CLI_NAME=redis-cli
REDIS_CLI_OBJ=anet.o adlist.o dict.o redis-cli.o zmalloc.o release.o ae.o crc64.o siphash.o crc16.o
REDIS_BENCHMARK_NAME=redis-benchmark
+40 -2
View File
@@ -31,6 +31,9 @@
*
* API:
*
* getTimeoutFromObjectOrReply() is just an utility function to parse a
* timeout argument since blocking operations usually require a timeout.
*
* blockClient() set the CLIENT_BLOCKED flag in the client, and set the
* specified block type 'btype' filed to one of BLOCKED_* macros.
*
@@ -64,6 +67,42 @@
int serveClientBlockedOnList(client *receiver, robj *key, robj *dstkey, redisDb *db, robj *value, int where);
/* Get a timeout value from an object and store it into 'timeout'.
* The final timeout is always stored as milliseconds as a time where the
* timeout will expire, however the parsing is performed according to
* the 'unit' that can be seconds or milliseconds.
*
* Note that if the timeout is zero (usually from the point of view of
* commands API this means no timeout) the value stored into 'timeout'
* is zero. */
int getTimeoutFromObjectOrReply(client *c, robj *object, mstime_t *timeout, int unit) {
long long tval;
long double ftval;
if (unit == UNIT_SECONDS) {
if (getLongDoubleFromObjectOrReply(c,object,&ftval,
"timeout is not an float or out of range") != C_OK)
return C_ERR;
tval = (long long) (ftval * 1000.0);
} else {
if (getLongLongFromObjectOrReply(c,object,&tval,
"timeout is not an integer or out of range") != C_OK)
return C_ERR;
}
if (tval < 0) {
addReplyError(c,"timeout is negative");
return C_ERR;
}
if (tval > 0) {
tval += mstime();
}
*timeout = tval;
return C_OK;
}
/* Block a client for the specific operation type. Once the CLIENT_BLOCKED
* flag is set client query buffer is not longer processed, but accumulated,
* and will be processed when the client is unblocked. */
@@ -72,7 +111,7 @@ void blockClient(client *c, int btype) {
c->btype = btype;
server.blocked_clients++;
server.blocked_clients_by_type[btype]++;
addClientToTimeoutTable(c);
addClientToShortTimeoutTable(c);
}
/* This function is called in the beforeSleep() function of the event loop
@@ -147,7 +186,6 @@ void unblockClient(client *c) {
server.blocked_clients_by_type[c->btype]--;
c->flags &= ~CLIENT_BLOCKED;
c->btype = BLOCKED_NONE;
removeClientFromTimeoutTable(c);
queueClientForReprocessing(c);
}
-3
View File
@@ -2755,9 +2755,6 @@ void replicationCacheMasterUsingMyself(void) {
delta);
server.master_initial_offset = server.master_repl_meaningful_offset;
server.repl_backlog_histlen -= delta;
server.repl_backlog_idx =
(server.repl_backlog_idx + (server.repl_backlog_size - delta)) %
server.repl_backlog_size;
if (server.repl_backlog_histlen < 0) server.repl_backlog_histlen = 0;
}
+135 -5
View File
@@ -1473,6 +1473,135 @@ int allPersistenceDisabled(void) {
return server.saveparamslen == 0 && server.aof_state == AOF_OFF;
}
/* ========================== Clients timeouts ============================= */
/* Check if this blocked client timedout (does nothing if the client is
* not blocked right now). If so send a reply, unblock it, and return 1.
* Otherwise 0 is returned and no operation is performed. */
int checkBlockedClientTimeout(client *c, mstime_t now) {
if (c->flags & CLIENT_BLOCKED &&
c->bpop.timeout != 0
&& c->bpop.timeout < now)
{
/* Handle blocking operation specific timeout. */
replyToBlockedClientTimedOut(c);
unblockClient(c);
return 1;
} else {
return 0;
}
}
/* Check for timeouts. Returns non-zero if the client was terminated.
* The function gets the current time in milliseconds as argument since
* it gets called multiple times in a loop, so calling gettimeofday() for
* each iteration would be costly without any actual gain. */
int clientsCronHandleTimeout(client *c, mstime_t now_ms) {
time_t now = now_ms/1000;
if (server.maxidletime &&
/* This handles the idle clients connection timeout if set. */
!(c->flags & CLIENT_SLAVE) && /* No timeout for slaves and monitors */
!(c->flags & CLIENT_MASTER) && /* No timeout for masters */
!(c->flags & CLIENT_BLOCKED) && /* No timeout for BLPOP */
!(c->flags & CLIENT_PUBSUB) && /* No timeout for Pub/Sub clients */
(now - c->lastinteraction > server.maxidletime))
{
serverLog(LL_VERBOSE,"Closing idle client");
freeClient(c);
return 1;
} else if (c->flags & CLIENT_BLOCKED) {
/* Blocked OPS timeout is handled with milliseconds resolution.
* However note that the actual resolution is limited by
* server.hz. So for short timeouts (less than SERVER_SHORT_TIMEOUT
* milliseconds) we populate a Radix tree and handle such timeouts
* in clientsHandleShortTimeout(). */
if (checkBlockedClientTimeout(c,now_ms)) return 0;
/* Cluster: handle unblock & redirect of clients blocked
* into keys no longer served by this server. */
if (server.cluster_enabled) {
if (clusterRedirectBlockedClientIfNeeded(c))
unblockClient(c);
}
}
return 0;
}
/* For shor timeouts, less than < CLIENT_SHORT_TIMEOUT milliseconds, we
* populate a radix tree of 128 bit keys composed as such:
*
* [8 byte big endian expire time]+[8 byte client ID]
*
* We don't do any cleanup in the Radix tree: when we run the clients that
* reached the timeout already, if they are no longer existing or no longer
* blocked with such timeout, we just go forward.
*
* Every time a client blocks with a short timeout, we add the client in
* the tree. In beforeSleep() we call clientsHandleShortTimeout() to run
* the tree and unblock the clients.
*
* Design hint: why we block only clients with short timeouts? For frugality:
* Clients blocking for 30 seconds usually don't need to be unblocked
* precisely, and anyway for the nature of Redis to *guarantee* unblock time
* precision is hard, so we can avoid putting a large number of clients in
* the radix tree without a good reason. This idea also has a role in memory
* usage as well given that we don't do cleanup, the shorter a client timeout,
* the less time it will stay in the radix tree. */
#define CLIENT_ST_KEYLEN 16 /* 8 bytes mstime + 8 bytes client ID. */
/* Given client ID and timeout, write the resulting radix tree key in buf. */
void encodeTimeoutKey(unsigned char *buf, uint64_t timeout, uint64_t id) {
timeout = htonu64(timeout);
memcpy(buf,&timeout,sizeof(timeout));
memcpy(buf+8,&id,sizeof(id));
}
/* Given a key encoded with encodeTimeoutKey(), resolve the fields and write
* the timeout into *toptr and the client ID into *idptr. */
void decodeTimeoutKey(unsigned char *buf, uint64_t *toptr, uint64_t *idptr) {
memcpy(toptr,buf,sizeof(*toptr));
*toptr = ntohu64(*toptr);
memcpy(idptr,buf+8,sizeof(*idptr));
}
/* Add the specified client id / timeout as a key in the radix tree we use
* to handle short timeouts. The client is not added to the list if its
* timeout is longer than CLIENT_SHORT_TIMEOUT milliseconds. */
void addClientToShortTimeoutTable(client *c) {
if (c->bpop.timeout == 0 ||
c->bpop.timeout - mstime() > CLIENT_SHORT_TIMEOUT)
{
return;
}
uint64_t timeout = c->bpop.timeout;
uint64_t id = c->id;
unsigned char buf[CLIENT_ST_KEYLEN];
encodeTimeoutKey(buf,timeout,id);
raxTryInsert(server.clients_timeout_table,buf,sizeof(buf),NULL,NULL);
}
/* This function is called in beforeSleep() in order to unblock ASAP clients
* that are waiting in blocking operations with a short timeout set. */
void clientsHandleShortTimeout(void) {
if (raxSize(server.clients_timeout_table) == 0) return;
uint64_t now = mstime();
raxIterator ri;
raxStart(&ri,server.clients_timeout_table);
raxSeek(&ri,"^",NULL,0);
while(raxNext(&ri)) {
uint64_t id, timeout;
decodeTimeoutKey(ri.key,&timeout,&id);
if (timeout >= now) break; /* All the timeouts are in the future. */
client *c = lookupClientByID(id);
if (c) checkBlockedClientTimeout(c,now);
raxRemove(server.clients_timeout_table,ri.key,ri.key_len,NULL);
raxSeek(&ri,"^",NULL,0);
}
}
/* ======================= Cron: called every 100 ms ======================== */
/* Add a sample to the operations per second array of samples. */
@@ -1618,6 +1747,9 @@ void getExpansiveClientsInfo(size_t *in_usage, size_t *out_usage) {
*/
#define CLIENTS_CRON_MIN_ITERATIONS 5
void clientsCron(void) {
/* Unblock short timeout clients ASAP. */
clientsHandleShortTimeout();
/* Try to process at least numclients/server.hz of clients
* per call. Since normally (if there are no big latency events) this
* function is called server.hz times per second, in the average case we
@@ -2057,7 +2189,7 @@ void beforeSleep(struct aeEventLoop *eventLoop) {
UNUSED(eventLoop);
/* Handle precise timeouts of blocked clients. */
handleBlockedClientsTimeout();
clientsHandleShortTimeout();
/* We should handle pending reads clients ASAP after event loop. */
handleClientsWithPendingReadsUsingThreads();
@@ -3969,13 +4101,11 @@ sds genRedisInfoString(const char *section) {
"client_recent_max_input_buffer:%zu\r\n"
"client_recent_max_output_buffer:%zu\r\n"
"blocked_clients:%d\r\n"
"tracking_clients:%d\r\n"
"clients_in_timeout_table:%lld\r\n",
"tracking_clients:%d\r\n",
listLength(server.clients)-listLength(server.slaves),
maxin, maxout,
server.blocked_clients,
server.tracking_clients,
raxSize(server.clients_timeout_table));
server.tracking_clients);
}
/* Memory */
+5 -8
View File
@@ -252,7 +252,6 @@ typedef long long ustime_t; /* microsecond time type. */
#define CLIENT_TRACKING_OPTOUT (1ULL<<35) /* Tracking in opt-out mode. */
#define CLIENT_TRACKING_CACHING (1ULL<<36) /* CACHING yes/no was given,
depending on optin/optout mode. */
#define CLIENT_IN_TO_TABLE (1ULL<<37) /* This client is in the timeout table. */
/* Client block type (btype field in client structure)
* if CLIENT_BLOCKED flag is set. */
@@ -278,6 +277,9 @@ typedef long long ustime_t; /* microsecond time type. */
buffer configuration. Just the first
three: normal, slave, pubsub. */
/* Other client related defines. */
#define CLIENT_SHORT_TIMEOUT 2000 /* See clientsHandleShortTimeout(). */
/* Slave replication state. Used in server.repl_state for slaves to remember
* what to do next. */
#define REPL_STATE_NONE 0 /* No active replication */
@@ -1068,7 +1070,7 @@ struct redisServer {
list *clients_pending_read; /* Client has pending read socket buffers. */
list *slaves, *monitors; /* List of slaves and MONITORs */
client *current_client; /* Current client executing the command. */
rax *clients_timeout_table; /* Radix tree for blocked clients timeouts. */
rax *clients_timeout_table; /* Radix tree for clients with short timeout. */
long fixed_time_expire; /* If > 0, expire keys against server.mstime. */
rax *clients_index; /* Active clients dictionary by client ID. */
int clients_paused; /* True if clients are currently paused */
@@ -2138,12 +2140,7 @@ void disconnectAllBlockedClients(void);
void handleClientsBlockedOnKeys(void);
void signalKeyAsReady(redisDb *db, robj *key);
void blockForKeys(client *c, int btype, robj **keys, int numkeys, mstime_t timeout, robj *target, streamID *ids);
/* timeout.c -- Blocked clients timeout and connections timeout. */
void addClientToTimeoutTable(client *c);
void removeClientFromTimeoutTable(client *c);
void handleBlockedClientsTimeout(void);
int clientsCronHandleTimeout(client *c, mstime_t now_ms);
void addClientToShortTimeoutTable(client *c);
/* expire.c -- Handling of expired keys */
void activeExpireCycle(int type);
-189
View File
@@ -1,189 +0,0 @@
/* Copyright (c) 2009-2020, Salvatore Sanfilippo <antirez at gmail dot com>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of Redis nor the names of its contributors may be used
* to endorse or promote products derived from this software without
* specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
#include "server.h"
#include "cluster.h"
/* ========================== Clients timeouts ============================= */
/* Check if this blocked client timedout (does nothing if the client is
* not blocked right now). If so send a reply, unblock it, and return 1.
* Otherwise 0 is returned and no operation is performed. */
int checkBlockedClientTimeout(client *c, mstime_t now) {
if (c->flags & CLIENT_BLOCKED &&
c->bpop.timeout != 0
&& c->bpop.timeout < now)
{
/* Handle blocking operation specific timeout. */
replyToBlockedClientTimedOut(c);
unblockClient(c);
return 1;
} else {
return 0;
}
}
/* Check for timeouts. Returns non-zero if the client was terminated.
* The function gets the current time in milliseconds as argument since
* it gets called multiple times in a loop, so calling gettimeofday() for
* each iteration would be costly without any actual gain. */
int clientsCronHandleTimeout(client *c, mstime_t now_ms) {
time_t now = now_ms/1000;
if (server.maxidletime &&
/* This handles the idle clients connection timeout if set. */
!(c->flags & CLIENT_SLAVE) && /* No timeout for slaves and monitors */
!(c->flags & CLIENT_MASTER) && /* No timeout for masters */
!(c->flags & CLIENT_BLOCKED) && /* No timeout for BLPOP */
!(c->flags & CLIENT_PUBSUB) && /* No timeout for Pub/Sub clients */
(now - c->lastinteraction > server.maxidletime))
{
serverLog(LL_VERBOSE,"Closing idle client");
freeClient(c);
return 1;
} else if (c->flags & CLIENT_BLOCKED) {
/* Cluster: handle unblock & redirect of clients blocked
* into keys no longer served by this server. */
if (server.cluster_enabled) {
if (clusterRedirectBlockedClientIfNeeded(c))
unblockClient(c);
}
}
return 0;
}
/* For blocked clients timeouts we populate a radix tree of 128 bit keys
* composed as such:
*
* [8 byte big endian expire time]+[8 byte client ID]
*
* We don't do any cleanup in the Radix tree: when we run the clients that
* reached the timeout already, if they are no longer existing or no longer
* blocked with such timeout, we just go forward.
*
* Every time a client blocks with a timeout, we add the client in
* the tree. In beforeSleep() we call handleBlockedClientsTimeout() to run
* the tree and unblock the clients. */
#define CLIENT_ST_KEYLEN 16 /* 8 bytes mstime + 8 bytes client ID. */
/* Given client ID and timeout, write the resulting radix tree key in buf. */
void encodeTimeoutKey(unsigned char *buf, uint64_t timeout, client *c) {
timeout = htonu64(timeout);
memcpy(buf,&timeout,sizeof(timeout));
memcpy(buf+8,&c,sizeof(c));
if (sizeof(c) == 4) memset(buf+12,0,4); /* Zero padding for 32bit target. */
}
/* Given a key encoded with encodeTimeoutKey(), resolve the fields and write
* the timeout into *toptr and the client pointer into *cptr. */
void decodeTimeoutKey(unsigned char *buf, uint64_t *toptr, client **cptr) {
memcpy(toptr,buf,sizeof(*toptr));
*toptr = ntohu64(*toptr);
memcpy(cptr,buf+8,sizeof(*cptr));
}
/* Add the specified client id / timeout as a key in the radix tree we use
* to handle blocked clients timeouts. The client is not added to the list
* if its timeout is zero (block forever). */
void addClientToTimeoutTable(client *c) {
if (c->bpop.timeout == 0) return;
uint64_t timeout = c->bpop.timeout;
unsigned char buf[CLIENT_ST_KEYLEN];
encodeTimeoutKey(buf,timeout,c);
if (raxTryInsert(server.clients_timeout_table,buf,sizeof(buf),NULL,NULL))
c->flags |= CLIENT_IN_TO_TABLE;
}
/* Remove the client from the table when it is unblocked for reasons
* different than timing out. */
void removeClientFromTimeoutTable(client *c) {
if (!(c->flags & CLIENT_IN_TO_TABLE)) return;
c->flags &= ~CLIENT_IN_TO_TABLE;
uint64_t timeout = c->bpop.timeout;
unsigned char buf[CLIENT_ST_KEYLEN];
encodeTimeoutKey(buf,timeout,c);
raxRemove(server.clients_timeout_table,buf,sizeof(buf),NULL);
}
/* This function is called in beforeSleep() in order to unblock clients
* that are waiting in blocking operations with a timeout set. */
void handleBlockedClientsTimeout(void) {
if (raxSize(server.clients_timeout_table) == 0) return;
uint64_t now = mstime();
raxIterator ri;
raxStart(&ri,server.clients_timeout_table);
raxSeek(&ri,"^",NULL,0);
while(raxNext(&ri)) {
uint64_t timeout;
client *c;
decodeTimeoutKey(ri.key,&timeout,&c);
if (timeout >= now) break; /* All the timeouts are in the future. */
c->flags &= ~CLIENT_IN_TO_TABLE;
checkBlockedClientTimeout(c,now);
raxRemove(server.clients_timeout_table,ri.key,ri.key_len,NULL);
raxSeek(&ri,"^",NULL,0);
}
}
/* Get a timeout value from an object and store it into 'timeout'.
* The final timeout is always stored as milliseconds as a time where the
* timeout will expire, however the parsing is performed according to
* the 'unit' that can be seconds or milliseconds.
*
* Note that if the timeout is zero (usually from the point of view of
* commands API this means no timeout) the value stored into 'timeout'
* is zero. */
int getTimeoutFromObjectOrReply(client *c, robj *object, mstime_t *timeout, int unit) {
long long tval;
long double ftval;
if (unit == UNIT_SECONDS) {
if (getLongDoubleFromObjectOrReply(c,object,&ftval,
"timeout is not an float or out of range") != C_OK)
return C_ERR;
tval = (long long) (ftval * 1000.0);
} else {
if (getLongLongFromObjectOrReply(c,object,&tval,
"timeout is not an integer or out of range") != C_OK)
return C_ERR;
}
if (tval < 0) {
addReplyError(c,"timeout is negative");
return C_ERR;
}
if (tval > 0) {
tval += mstime();
}
*timeout = tval;
return C_OK;
}