perf: performance improvements (#107-#111)

Reviewed-on: #86

src/client/client_send.c

@@ -21,6 +21,9 @@
 #include <string.h>
 #include <threads.h>
 #include <time.h>
+#include <unistd.h>
+
+#define STREAM_THRESHOLD (64ULL * 1024 * 1024)
 
 static int incremental_check(Client* client, File* file, DeltaSignature** out_sig) {
   *out_sig = NULL;
@@ -208,10 +211,13 @@ int send_chunk(Client* client, Chunk* chunk, Config* config) {
     return 0;
   }
 
-  bool use_sendfile = config->use_sendfile && !config->use_compression;
   for (int i = 0; i < chunk->element_count; i++) {
-    int rc =
+    File* f = chunk->items[i];
-        send_single_file(client, chunk->items[i], config, config->use_incremental, use_sendfile);
+    if (f == NULL)
+      continue;
+    bool stream = f->data->data == NULL && f->data->size > 0;
+    bool use_sendfile = (config->use_sendfile && !config->use_compression) || stream;
+    int rc = send_single_file(client, f, config, config->use_incremental, use_sendfile);
     if (rc == 1)
       continue;
     if (rc < 0)
@@ -295,16 +301,14 @@ static int send_chunks_multithreaded(void* pipeline_context) {
 
 static int scan_directory_multithreaded(void* pipeline_context) {
   PipelineContextSender* context = (PipelineContextSender*)pipeline_context;
-  mtx_lock(&context->mutex_scanner);
+  ParallelScanner* scanner = parallel_scanner_create(
-  DirectoryScanner* scanner = directory_scanner_create(
       context->config->send_directory, context->config->use_metadata, context->config->chunk_size,
       context->config->exclude_patterns, context->config->exclude_count,
       context->config->include_patterns, context->config->include_count, context->config->max_size,
-      context->config->min_size);
+      context->config->min_size, 4);
-  mtx_unlock(&context->mutex_scanner);
 
   Chunk* current_chunk;
-  while ((current_chunk = directory_scanner_next(scanner)) != NULL) {
+  while ((current_chunk = parallel_scanner_next(scanner)) != NULL) {
     if (context->config->use_delete) {
       mtx_lock(&context->mutex_scanner);
       for (int i = 0; i < current_chunk->element_count; i++) {
@@ -324,7 +328,7 @@ static int scan_directory_multithreaded(void* pipeline_context) {
   cnd_signal(&context->condition_not_empty_scanner);
   mtx_unlock(&context->mutex_scanner);
 
-  directory_scanner_destroy(scanner);
+  parallel_scanner_destroy(scanner);
   return thrd_success;
 }
 
@@ -343,9 +347,12 @@ static int load_files_multithreaded(void* pipeline_context) {
     }
     if (!context->config->use_sendfile) {
       for (int i = 0; i < chunk->element_count; i++) {
-        if (!file_load_data(chunk->items[i])) {
+        File* f = chunk->items[i];
+        if (f->data->size > STREAM_THRESHOLD)
+          continue;
+        if (!file_load_data(f)) {
           log_message(LOG_LEVEL_ERROR, "Failed to load file data, skipping");
-          file_destroy(chunk->items[i]);
+          file_destroy(f);
           chunk->items[i] = NULL;
         }
       }
@@ -433,7 +440,10 @@ int send_files(Config* config) {
     }
     if (!config->use_sendfile) {
       for (int i = 0; i < current_chunk->element_count; i++) {
-        if (!file_load_data(current_chunk->items[i])) {
+        File* f = current_chunk->items[i];
+        if (f->data->size > STREAM_THRESHOLD)
+          continue;
+        if (!file_load_data(f)) {
           log_message(LOG_LEVEL_ERROR, "Failed to load file data");
           continue;
         }
@@ -518,8 +528,20 @@ int send_files_multithreaded(Config* config) {
     return 0;
   }
 
-  Queue* q1 = queue_create(100, chunk_destroy);
+  long pages = sysconf(_SC_AVPHYS_PAGES);
-  Queue* q2 = queue_create(100, chunk_destroy);
+  long page_size = sysconf(_SC_PAGE_SIZE);
+  unsigned long long available_memory =
+      pages > 0 && page_size > 0 ? (unsigned long long)pages * (unsigned long long)page_size
+                                 : 512ULL * 1024 * 1024;
+  unsigned long long avg_file_size = 1024 * 1024;
+  int qsize = (int)(available_memory / avg_file_size);
+  if (qsize < 10)
+    qsize = 10;
+  if (qsize > 1000)
+    qsize = 1000;
+
+  Queue* q1 = queue_create(qsize, chunk_destroy);
+  Queue* q2 = queue_create(qsize, chunk_destroy);
   if (!q1 || !q2) {
     if (q1)
       queue_destroy(q1);

src/client/scanner.c

@@ -9,6 +9,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/stat.h>
+#include <threads.h>
 #include <unistd.h>
 
 DirectoryScanner* directory_scanner_create(char* root_directory, bool use_metadata,
@@ -55,7 +56,6 @@ static Chunk* chunk_data_to_chunk(ArrayList* chunk_data) {
   return chunk;
 }
 
-// Returns: 1 on success, 0 if no more directories in queue, -1 on opendir failure
 static int open_next_directory(DirectoryScanner* scanner) {
   if (scanner->current_dir) {
     closedir(scanner->current_dir);
@@ -167,3 +167,253 @@ Chunk* directory_scanner_next(DirectoryScanner* scanner) {
   array_list_delete(chunk_data);
   return NULL;
 }
+
+typedef struct {
+  ParallelScanner* ps;
+  char** dirs;
+  int dir_count;
+  bool use_metadata;
+  unsigned long long chunk_size;
+  char** exclude_patterns;
+  int exclude_count;
+  char** include_patterns;
+  int include_count;
+  unsigned long long max_size;
+  unsigned long long min_size;
+} ParallelWorkerArg;
+
+static int parallel_worker_thread(void* arg) {
+  ParallelWorkerArg* wa = (ParallelWorkerArg*)arg;
+  for (int i = 0; i < wa->dir_count; i++) {
+    DirectoryScanner* ds = directory_scanner_create(
+        wa->dirs[i], wa->use_metadata, wa->chunk_size, wa->exclude_patterns, wa->exclude_count,
+        wa->include_patterns, wa->include_count, wa->max_size, wa->min_size);
+    Chunk* chunk;
+    while ((chunk = directory_scanner_next(ds)) != NULL) {
+      queue_enqueue_multithreaded(wa->ps->result_queue, chunk, &wa->ps->result_mutex,
+                                  &wa->ps->result_not_empty, &wa->ps->result_not_full);
+    }
+    directory_scanner_destroy(ds);
+    free(wa->dirs[i]);
+  }
+  ParallelScanner* ps = wa->ps;
+  free(wa->dirs);
+  free(wa);
+  mtx_lock(&ps->result_mutex);
+  ps->completed++;
+  if (ps->completed >= ps->num_threads) {
+    ps->done = true;
+    cnd_signal(&ps->result_not_empty);
+  }
+  mtx_unlock(&ps->result_mutex);
+  return thrd_success;
+}
+
+ParallelScanner* parallel_scanner_create(char* root_directory, bool use_metadata,
+                                         unsigned long long chunk_size, char** exclude_patterns,
+                                         int exclude_count, char** include_patterns,
+                                         int include_count, unsigned long long max_size,
+                                         unsigned long long min_size, int num_threads) {
+  ParallelScanner* ps = calloc(1, sizeof(ParallelScanner));
+  if (!ps)
+    return NULL;
+  ps->result_queue = queue_create(100, chunk_destroy);
+  if (!ps->result_queue) {
+    free(ps);
+    return NULL;
+  }
+  if (mtx_init(&ps->result_mutex, mtx_plain) != thrd_success ||
+      cnd_init(&ps->result_not_empty) != thrd_success ||
+      cnd_init(&ps->result_not_full) != thrd_success) {
+    queue_destroy(ps->result_queue);
+    free(ps);
+    return NULL;
+  }
+
+  DIR* dir = opendir(root_directory);
+  if (!dir) {
+    perror("Could not open root directory for parallel scan");
+    parallel_scanner_destroy(ps);
+    return NULL;
+  }
+
+  ArrayList* root_files = array_list_create(file_destroy);
+  ArrayList* subdirs = array_list_create(free);
+  struct dirent* entry;
+  while ((entry = readdir(dir)) != NULL) {
+    if (strcmp(entry->d_name, ".") == 0 || strcmp(entry->d_name, "..") == 0)
+      continue;
+    char* cur_path = path_cat(root_directory, entry->d_name);
+    if (!cur_path)
+      continue;
+    struct stat st;
+    if (stat(cur_path, &st) != 0) {
+      free(cur_path);
+      continue;
+    }
+    if (S_ISDIR(st.st_mode)) {
+      array_list_add(subdirs, cur_path);
+    } else {
+      bool excluded = false;
+      for (int i = 0; i < exclude_count; i++) {
+        if (glob_match(exclude_patterns[i], entry->d_name)) {
+          excluded = true;
+          break;
+        }
+      }
+      if (excluded) {
+        free(cur_path);
+        continue;
+      }
+      if (include_count > 0) {
+        bool included = false;
+        for (int i = 0; i < include_count; i++) {
+          if (glob_match(include_patterns[i], entry->d_name)) {
+            included = true;
+            break;
+          }
+        }
+        if (!included) {
+          free(cur_path);
+          continue;
+        }
+      }
+      if ((max_size > 0 && (unsigned long long)st.st_size > max_size) ||
+          (min_size > 0 && (unsigned long long)st.st_size < min_size)) {
+        free(cur_path);
+        continue;
+      }
+      File* file = file_create(cur_path);
+      free(cur_path);
+      if (!file)
+        continue;
+      file->data->size = st.st_size;
+      if (use_metadata)
+        file->metadata = file_metadata_create(&st);
+      array_list_add(root_files, file);
+    }
+  }
+  closedir(dir);
+
+  unsigned long long cs = chunk_size > 0 ? chunk_size : DESIRED_CHUNK_SIZE;
+  if (root_files->size > 0) {
+    ArrayList* batch = array_list_create(NULL);
+    unsigned long long batch_size = 0;
+    Chunk* first = NULL;
+    for (int i = 0; i < root_files->size; i++) {
+      File* f = (File*)root_files->items[i];
+      array_list_add(batch, f);
+      batch_size += f->data->size;
+      if (batch_size >= cs || i == root_files->size - 1) {
+        void** items = array_list_to_array(batch);
+        Chunk* c = chunk_create((File**)items, batch->size);
+        free(items);
+        batch->item_destroyer = NULL;
+        array_list_delete(batch);
+        batch = NULL;
+        if (!first) {
+          first = c;
+        } else {
+          queue_enqueue_multithreaded(ps->result_queue, c, &ps->result_mutex, &ps->result_not_empty,
+                                      &ps->result_not_full);
+        }
+        if (i < root_files->size - 1) {
+          batch = array_list_create(NULL);
+          batch_size = 0;
+        }
+      }
+    }
+    if (batch) {
+      batch->item_destroyer = NULL;
+      array_list_delete(batch);
+    }
+    ps->initial_chunk = first;
+    root_files->item_destroyer = NULL;
+  }
+  array_list_delete(root_files);
+
+  int n = num_threads > 0 ? num_threads : 4;
+  if (n > subdirs->size)
+    n = subdirs->size > 0 ? subdirs->size : 1;
+
+  if (subdirs->size > 0) {
+    ps->num_threads = n;
+    ps->threads = calloc(n, sizeof(thrd_t));
+    if (!ps->threads) {
+      array_list_delete(subdirs);
+      parallel_scanner_destroy(ps);
+      return NULL;
+    }
+    int dirs_per_thread = subdirs->size / n;
+    int remainder = subdirs->size % n;
+    int start = 0;
+    for (int t = 0; t < n; t++) {
+      int count = dirs_per_thread + (t < remainder ? 1 : 0);
+      if (count == 0)
+        break;
+      ParallelWorkerArg* wa = calloc(1, sizeof(ParallelWorkerArg));
+      if (!wa)
+        break;
+      wa->ps = ps;
+      wa->dirs = calloc(count, sizeof(char*));
+      if (!wa->dirs) {
+        free(wa);
+        break;
+      }
+      for (int j = 0; j < count; j++)
+        wa->dirs[j] = str_dup((char*)subdirs->items[start + j]);
+      wa->dir_count = count;
+      wa->use_metadata = use_metadata;
+      wa->chunk_size = cs;
+      wa->exclude_patterns = exclude_patterns;
+      wa->exclude_count = exclude_count;
+      wa->include_patterns = include_patterns;
+      wa->include_count = include_count;
+      wa->max_size = max_size;
+      wa->min_size = min_size;
+      start += count;
+      if (thrd_create(&ps->threads[t], parallel_worker_thread, wa) != thrd_success) {
+        for (int j = 0; j < count; j++)
+          free(wa->dirs[j]);
+        free(wa->dirs);
+        free(wa);
+        ps->num_threads = t;
+        break;
+      }
+    }
+  }
+  array_list_delete(subdirs);
+  return ps;
+}
+
+Chunk* parallel_scanner_next(ParallelScanner* ps) {
+  if (ps->initial_chunk) {
+    Chunk* c = ps->initial_chunk;
+    ps->initial_chunk = NULL;
+    return c;
+  }
+  if (ps->num_threads == 0) {
+    ps->done = true;
+    return NULL;
+  }
+  Chunk* chunk = queue_dequeue_multithreaded(
+      ps->result_queue, &ps->result_mutex, &ps->result_not_empty, &ps->result_not_full, &ps->done);
+  return chunk;
+}
+
+void parallel_scanner_destroy(ParallelScanner* ps) {
+  if (!ps)
+    return;
+  ps->done = true;
+  cnd_signal(&ps->result_not_empty);
+  for (int i = 0; i < ps->num_threads; i++)
+    thrd_join(ps->threads[i], NULL);
+  free(ps->threads);
+  if (ps->initial_chunk)
+    chunk_destroy(ps->initial_chunk);
+  queue_destroy(ps->result_queue);
+  mtx_destroy(&ps->result_mutex);
+  cnd_destroy(&ps->result_not_empty);
+  cnd_destroy(&ps->result_not_full);
+  free(ps);
+}

src/client/scanner.h

@@ -5,6 +5,7 @@
 #include "queue.h"
 #include <dirent.h>
 #include <stdbool.h>
+#include <threads.h>
 
 typedef struct {
   Queue* directories;
@@ -20,6 +21,18 @@ typedef struct {
   unsigned long long min_size;
 } DirectoryScanner;
 
+typedef struct {
+  Queue* result_queue;
+  mtx_t result_mutex;
+  cnd_t result_not_empty;
+  cnd_t result_not_full;
+  int num_threads;
+  thrd_t* threads;
+  bool done;
+  int completed;
+  Chunk* initial_chunk;
+} ParallelScanner;
+
 DirectoryScanner* directory_scanner_create(char* root_directory, bool use_metadata,
                                            unsigned long long chunk_size, char** exclude_patterns,
                                            int exclude_count, char** include_patterns,
@@ -28,4 +41,12 @@ DirectoryScanner* directory_scanner_create(char* root_directory, bool use_metada
 Chunk* directory_scanner_next(DirectoryScanner* scanner);
 void directory_scanner_destroy(DirectoryScanner* scanner);
 
+ParallelScanner* parallel_scanner_create(char* root_directory, bool use_metadata,
+                                         unsigned long long chunk_size, char** exclude_patterns,
+                                         int exclude_count, char** include_patterns,
+                                         int include_count, unsigned long long max_size,
+                                         unsigned long long min_size, int num_threads);
+Chunk* parallel_scanner_next(ParallelScanner* scanner);
+void parallel_scanner_destroy(ParallelScanner* scanner);
+
 #endif

src/shared/compression.c

@@ -2,10 +2,28 @@
 #include "data.h"
 #include "log.h"
 #include "stdlib.h"
+#include "string.h"
+#include <strings.h>
 #include "zstd.h"
 
 #define INITIAL_DECOMPRESS_BUF_SIZE (1024 * 1024)
 
+static const char* SKIP_COMPRESSION_EXTENSIONS[] = {".jpg", ".jpeg", ".png", ".gif", ".mp4", ".mkv",
+                                                    ".zip", ".gz",   ".xz",  ".zst", NULL};
+
+bool compression_should_skip(const char* path) {
+  if (!path)
+    return false;
+  const char* dot = strrchr(path, '.');
+  if (!dot)
+    return false;
+  for (int i = 0; SKIP_COMPRESSION_EXTENSIONS[i]; i++) {
+    if (strcasecmp(dot, SKIP_COMPRESSION_EXTENSIONS[i]) == 0)
+      return true;
+  }
+  return false;
+}
+
 Data* data_compress(Data* data_to_compress, int compression_level) {
   log_message(LOG_LEVEL_DEBUG, "Starting to compress data");
   size_t dst_size = ZSTD_compressBound(data_to_compress->size);

src/shared/compression.h

@@ -2,8 +2,10 @@
 #define COMPRESSION_H
 
 #include "data.h"
+#include <stdbool.h>
 
 Data* data_compress(Data* data_to_compress, int compression_level);
 Data* data_decompress(Data* compressed_data);
+bool compression_should_skip(const char* path);
 
 #endif

src/shared/file.c

@@ -102,7 +102,7 @@ bool file_send_single_calls(File* file, int file_descriptor, bool use_metadata,
                             int compression_level, bool send_path) {
   const Data* data_to_send = file->data;
   Data* compressed_data = NULL;
-  if (compression_level > 0) {
+  if (compression_level > 0 && !compression_should_skip(file->path)) {
     compressed_data = data_compress(file->data, compression_level);
     if (compressed_data == NULL) {
       log_message(LOG_LEVEL_ERROR, "Failed to compress file data");

src/shared/protocol.c

@@ -2,6 +2,7 @@
 #include "log.h"
 #include <errno.h>
 #include <openssl/ssl.h>
+#include <poll.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -10,7 +11,7 @@
 
 static __thread int io_read_fd = -1;
 static __thread int io_write_fd = -1;
-static SSL* io_ssl = NULL;
+static SSL* io_ssl;
 
 static unsigned long long io_bwlimit = 0;
 static long long bw_tokens = 0;
@@ -46,12 +47,11 @@ static void bw_throttle(size_t bytes_written) {
   bw_tokens -= (long long)bytes_written;
 
   if (bw_tokens < 0) {
-    long long deficit_ns = (long long)((double)(-bw_tokens) / io_bwlimit * 1000000000.0);
+    long long deficit_us = (long long)((double)(-bw_tokens) / io_bwlimit * 1000000.0);
-    struct timespec sleep_time, remaining;
+    if (deficit_us >= 1000)
-    sleep_time.tv_sec = deficit_ns / 1000000000LL;
+      poll(NULL, 0, (int)(deficit_us / 1000));
-    sleep_time.tv_nsec = deficit_ns % 1000000000LL;
+    else
-    while (nanosleep(&sleep_time, &remaining) < 0 && errno == EINTR)
+      usleep((useconds_t)deficit_us);
-      sleep_time = remaining;
     bw_tokens = 0;
     clock_gettime(CLOCK_MONOTONIC, &bw_last_refill);
   }

src/shared/protocol.h

@@ -17,7 +17,10 @@ enum NET_STATUS {
   STATUS_MANIFEST,
   STATUS_CHECK,
   STATUS_DELTA_SIGNATURE,
-  STATUS_DELTA_DATA
+  STATUS_DELTA_DATA,
+  STATUS_KEEPALIVE,
+  STATUS_ABORT,
+  STATUS_CHECK_BATCH
 };
 
 void io_set_fds(int read_fd, int write_fd);