xref: /f-stack/app/redis-5.0.5/src/bio.c (revision 572c4311)
1 /* Background I/O service for Redis.
2  *
3  * This file implements operations that we need to perform in the background.
4  * Currently there is only a single operation, that is a background close(2)
5  * system call. This is needed as when the process is the last owner of a
6  * reference to a file closing it means unlinking it, and the deletion of the
7  * file is slow, blocking the server.
8  *
9  * In the future we'll either continue implementing new things we need or
10  * we'll switch to libeio. However there are probably long term uses for this
11  * file as we may want to put here Redis specific background tasks (for instance
12  * it is not impossible that we'll need a non blocking FLUSHDB/FLUSHALL
13  * implementation).
14  *
15  * DESIGN
16  * ------
17  *
18  * The design is trivial, we have a structure representing a job to perform
19  * and a different thread and job queue for every job type.
20  * Every thread waits for new jobs in its queue, and process every job
21  * sequentially.
22  *
23  * Jobs of the same type are guaranteed to be processed from the least
24  * recently inserted to the most recently inserted (older jobs processed
25  * first).
26  *
27  * Currently there is no way for the creator of the job to be notified about
28  * the completion of the operation, this will only be added when/if needed.
29  *
30  * ----------------------------------------------------------------------------
31  *
32  * Copyright (c) 2009-2012, Salvatore Sanfilippo <antirez at gmail dot com>
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions are met:
37  *
38  *   * Redistributions of source code must retain the above copyright notice,
39  *     this list of conditions and the following disclaimer.
40  *   * Redistributions in binary form must reproduce the above copyright
41  *     notice, this list of conditions and the following disclaimer in the
42  *     documentation and/or other materials provided with the distribution.
43  *   * Neither the name of Redis nor the names of its contributors may be used
44  *     to endorse or promote products derived from this software without
45  *     specific prior written permission.
46  *
47  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
48  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
50  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
51  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
52  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
53  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
54  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
55  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
56  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
57  * POSSIBILITY OF SUCH DAMAGE.
58  */
59 
60 
61 #include "server.h"
62 #include "bio.h"
63 
64 static pthread_t bio_threads[BIO_NUM_OPS];
65 static pthread_mutex_t bio_mutex[BIO_NUM_OPS];
66 static pthread_cond_t bio_newjob_cond[BIO_NUM_OPS];
67 static pthread_cond_t bio_step_cond[BIO_NUM_OPS];
68 static list *bio_jobs[BIO_NUM_OPS];
69 /* The following array is used to hold the number of pending jobs for every
70  * OP type. This allows us to export the bioPendingJobsOfType() API that is
71  * useful when the main thread wants to perform some operation that may involve
72  * objects shared with the background thread. The main thread will just wait
73  * that there are no longer jobs of this type to be executed before performing
74  * the sensible operation. This data is also useful for reporting. */
75 static unsigned long long bio_pending[BIO_NUM_OPS];
76 
77 /* This structure represents a background Job. It is only used locally to this
78  * file as the API does not expose the internals at all. */
79 struct bio_job {
80     time_t time; /* Time at which the job was created. */
81     /* Job specific arguments pointers. If we need to pass more than three
82      * arguments we can just pass a pointer to a structure or alike. */
83     void *arg1, *arg2, *arg3;
84 };
85 
86 void *bioProcessBackgroundJobs(void *arg);
87 void lazyfreeFreeObjectFromBioThread(robj *o);
88 void lazyfreeFreeDatabaseFromBioThread(dict *ht1, dict *ht2);
89 void lazyfreeFreeSlotsMapFromBioThread(zskiplist *sl);
90 
91 /* Make sure we have enough stack to perform all the things we do in the
92  * main thread. */
93 #define REDIS_THREAD_STACK_SIZE (1024*1024*4)
94 
95 /* Initialize the background system, spawning the thread. */
bioInit(void)96 void bioInit(void) {
97     pthread_attr_t attr;
98     pthread_t thread;
99     size_t stacksize;
100     int j;
101 
102     /* Initialization of state vars and objects */
103     for (j = 0; j < BIO_NUM_OPS; j++) {
104         pthread_mutex_init(&bio_mutex[j],NULL);
105         pthread_cond_init(&bio_newjob_cond[j],NULL);
106         pthread_cond_init(&bio_step_cond[j],NULL);
107         bio_jobs[j] = listCreate();
108         bio_pending[j] = 0;
109     }
110 
111     /* Set the stack size as by default it may be small in some system */
112     pthread_attr_init(&attr);
113     pthread_attr_getstacksize(&attr,&stacksize);
114     if (!stacksize) stacksize = 1; /* The world is full of Solaris Fixes */
115     while (stacksize < REDIS_THREAD_STACK_SIZE) stacksize *= 2;
116     pthread_attr_setstacksize(&attr, stacksize);
117 
118     /* Ready to spawn our threads. We use the single argument the thread
119      * function accepts in order to pass the job ID the thread is
120      * responsible of. */
121     for (j = 0; j < BIO_NUM_OPS; j++) {
122         void *arg = (void*)(unsigned long) j;
123         if (pthread_create(&thread,&attr,bioProcessBackgroundJobs,arg) != 0) {
124             serverLog(LL_WARNING,"Fatal: Can't initialize Background Jobs.");
125             exit(1);
126         }
127         bio_threads[j] = thread;
128     }
129 }
130 
bioCreateBackgroundJob(int type,void * arg1,void * arg2,void * arg3)131 void bioCreateBackgroundJob(int type, void *arg1, void *arg2, void *arg3) {
132     struct bio_job *job = zmalloc(sizeof(*job));
133 
134     job->time = time(NULL);
135     job->arg1 = arg1;
136     job->arg2 = arg2;
137     job->arg3 = arg3;
138     pthread_mutex_lock(&bio_mutex[type]);
139     listAddNodeTail(bio_jobs[type],job);
140     bio_pending[type]++;
141     pthread_cond_signal(&bio_newjob_cond[type]);
142     pthread_mutex_unlock(&bio_mutex[type]);
143 }
144 
bioProcessBackgroundJobs(void * arg)145 void *bioProcessBackgroundJobs(void *arg) {
146     struct bio_job *job;
147     unsigned long type = (unsigned long) arg;
148     sigset_t sigset;
149 
150     /* Check that the type is within the right interval. */
151     if (type >= BIO_NUM_OPS) {
152         serverLog(LL_WARNING,
153             "Warning: bio thread started with wrong type %lu",type);
154         return NULL;
155     }
156 
157     /* Make the thread killable at any time, so that bioKillThreads()
158      * can work reliably. */
159     pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL);
160     pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, NULL);
161 
162     pthread_mutex_lock(&bio_mutex[type]);
163     /* Block SIGALRM so we are sure that only the main thread will
164      * receive the watchdog signal. */
165     sigemptyset(&sigset);
166     sigaddset(&sigset, SIGALRM);
167     if (pthread_sigmask(SIG_BLOCK, &sigset, NULL))
168         serverLog(LL_WARNING,
169             "Warning: can't mask SIGALRM in bio.c thread: %s", strerror(errno));
170 
171     while(1) {
172         listNode *ln;
173 
174         /* The loop always starts with the lock hold. */
175         if (listLength(bio_jobs[type]) == 0) {
176             pthread_cond_wait(&bio_newjob_cond[type],&bio_mutex[type]);
177             continue;
178         }
179         /* Pop the job from the queue. */
180         ln = listFirst(bio_jobs[type]);
181         job = ln->value;
182         /* It is now possible to unlock the background system as we know have
183          * a stand alone job structure to process.*/
184         pthread_mutex_unlock(&bio_mutex[type]);
185 
186         /* Process the job accordingly to its type. */
187         if (type == BIO_CLOSE_FILE) {
188             close((long)job->arg1);
189         } else if (type == BIO_AOF_FSYNC) {
190             redis_fsync((long)job->arg1);
191         } else if (type == BIO_LAZY_FREE) {
192             /* What we free changes depending on what arguments are set:
193              * arg1 -> free the object at pointer.
194              * arg2 & arg3 -> free two dictionaries (a Redis DB).
195              * only arg3 -> free the skiplist. */
196             if (job->arg1)
197                 lazyfreeFreeObjectFromBioThread(job->arg1);
198             else if (job->arg2 && job->arg3)
199                 lazyfreeFreeDatabaseFromBioThread(job->arg2,job->arg3);
200             else if (job->arg3)
201                 lazyfreeFreeSlotsMapFromBioThread(job->arg3);
202         } else {
203             serverPanic("Wrong job type in bioProcessBackgroundJobs().");
204         }
205         zfree(job);
206 
207         /* Lock again before reiterating the loop, if there are no longer
208          * jobs to process we'll block again in pthread_cond_wait(). */
209         pthread_mutex_lock(&bio_mutex[type]);
210         listDelNode(bio_jobs[type],ln);
211         bio_pending[type]--;
212 
213         /* Unblock threads blocked on bioWaitStepOfType() if any. */
214         pthread_cond_broadcast(&bio_step_cond[type]);
215     }
216 }
217 
218 /* Return the number of pending jobs of the specified type. */
bioPendingJobsOfType(int type)219 unsigned long long bioPendingJobsOfType(int type) {
220     unsigned long long val;
221     pthread_mutex_lock(&bio_mutex[type]);
222     val = bio_pending[type];
223     pthread_mutex_unlock(&bio_mutex[type]);
224     return val;
225 }
226 
227 /* If there are pending jobs for the specified type, the function blocks
228  * and waits that the next job was processed. Otherwise the function
229  * does not block and returns ASAP.
230  *
231  * The function returns the number of jobs still to process of the
232  * requested type.
233  *
234  * This function is useful when from another thread, we want to wait
235  * a bio.c thread to do more work in a blocking way.
236  */
bioWaitStepOfType(int type)237 unsigned long long bioWaitStepOfType(int type) {
238     unsigned long long val;
239     pthread_mutex_lock(&bio_mutex[type]);
240     val = bio_pending[type];
241     if (val != 0) {
242         pthread_cond_wait(&bio_step_cond[type],&bio_mutex[type]);
243         val = bio_pending[type];
244     }
245     pthread_mutex_unlock(&bio_mutex[type]);
246     return val;
247 }
248 
249 /* Kill the running bio threads in an unclean way. This function should be
250  * used only when it's critical to stop the threads for some reason.
251  * Currently Redis does this only on crash (for instance on SIGSEGV) in order
252  * to perform a fast memory check without other threads messing with memory. */
bioKillThreads(void)253 void bioKillThreads(void) {
254     int err, j;
255 
256     for (j = 0; j < BIO_NUM_OPS; j++) {
257         if (pthread_cancel(bio_threads[j]) == 0) {
258             if ((err = pthread_join(bio_threads[j],NULL)) != 0) {
259                 serverLog(LL_WARNING,
260                     "Bio thread for job type #%d can be joined: %s",
261                         j, strerror(err));
262             } else {
263                 serverLog(LL_WARNING,
264                     "Bio thread for job type #%d terminated",j);
265             }
266         }
267     }
268 }
269