chiark / gitweb /
journal: size journal data hash table based on maximum file size metrics
authorLennart Poettering <lennart@poettering.net>
Mon, 16 Jul 2012 22:36:15 +0000 (00:36 +0200)
committerLennart Poettering <lennart@poettering.net>
Mon, 16 Jul 2012 22:59:03 +0000 (00:59 +0200)
The default of 2047 hash table entries turned out to result in way too
many collisions for bigger files, hence scale the hash table size by the
estimated maximum file size.

src/journal/journal-file.c
src/journal/journal-file.h
src/journal/journald.c
src/journal/sd-journal.c
src/journal/test-journal-stream.c
src/journal/test-journal.c

index 9d8c9e6..04564d9 100644 (file)
@@ -32,8 +32,8 @@
 #include "lookup3.h"
 #include "compress.h"
 
-#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*16ULL)
-#define DEFAULT_FIELD_HASH_TABLE_SIZE (2047ULL*16ULL)
+#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
+#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
 
 #define DEFAULT_WINDOW_SIZE (8ULL*1024ULL*1024ULL)
 
@@ -509,7 +509,17 @@ static int journal_file_setup_data_hash_table(JournalFile *f) {
 
         assert(f);
 
-        s = DEFAULT_DATA_HASH_TABLE_SIZE;
+        /* We estimate that we need 1 hash table entry per 2K of
+           journal file and we want to make sure we never get beyond
+           75% fill level. Calculate the hash table size for the
+           maximum file size based on these metrics. */
+
+        s = (f->metrics.max_size * 4 / 2048 / 3) * sizeof(HashItem);
+        if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
+                s = DEFAULT_DATA_HASH_TABLE_SIZE;
+
+        log_info("Reserving %llu entries in hash table.", (unsigned long long) s);
+
         r = journal_file_append_object(f,
                                        OBJECT_DATA_HASH_TABLE,
                                        offsetof(Object, hash_table.items) + s,
@@ -1939,6 +1949,7 @@ int journal_file_open(
                 const char *fname,
                 int flags,
                 mode_t mode,
+                JournalMetrics *metrics,
                 JournalFile *template,
                 JournalFile **ret) {
 
@@ -1965,10 +1976,8 @@ int journal_file_open(
         f->writable = (flags & O_ACCMODE) != O_RDONLY;
         f->prot = prot_from_flags(flags);
 
-        if (template) {
-                f->metrics = template->metrics;
+        if (template)
                 f->compress = template->compress;
-        }
 
         f->path = strdup(fname);
         if (!f->path) {
@@ -2019,6 +2028,12 @@ int journal_file_open(
         }
 
         if (f->writable) {
+                if (metrics) {
+                        journal_default_metrics(metrics, f->fd);
+                        f->metrics = *metrics;
+                } else if (template)
+                        f->metrics = template->metrics;
+
                 r = journal_file_refresh_header(f);
                 if (r < 0)
                         goto fail;
@@ -2093,7 +2108,7 @@ int journal_file_rotate(JournalFile **f) {
 
         old_file->header->state = STATE_ARCHIVED;
 
-        r = journal_file_open(old_file->path, old_file->flags, old_file->mode, old_file, &new_file);
+        r = journal_file_open(old_file->path, old_file->flags, old_file->mode, NULL, old_file, &new_file);
         journal_file_close(old_file);
 
         *f = new_file;
@@ -2104,6 +2119,7 @@ int journal_file_open_reliably(
                 const char *fname,
                 int flags,
                 mode_t mode,
+                JournalMetrics *metrics,
                 JournalFile *template,
                 JournalFile **ret) {
 
@@ -2111,7 +2127,7 @@ int journal_file_open_reliably(
         size_t l;
         char *p;
 
-        r = journal_file_open(fname, flags, mode, template, ret);
+        r = journal_file_open(fname, flags, mode, metrics, template, ret);
         if (r != -EBADMSG && /* corrupted */
             r != -ENODATA && /* truncated */
             r != -EHOSTDOWN && /* other machine */
@@ -2140,7 +2156,7 @@ int journal_file_open_reliably(
 
         log_warning("File %s corrupted, renaming and replacing.", fname);
 
-        return journal_file_open(fname, flags, mode, template, ret);
+        return journal_file_open(fname, flags, mode, metrics, template, ret);
 }
 
 struct vacuum_info {
index 6219742..1f7f73a 100644 (file)
@@ -88,10 +88,23 @@ typedef enum direction {
         DIRECTION_DOWN
 } direction_t;
 
-int journal_file_open(const char *fname, int flags, mode_t mode, JournalFile *template, JournalFile **ret);
+int journal_file_open(
+                const char *fname,
+                int flags,
+                mode_t mode,
+                JournalMetrics *metrics,
+                JournalFile *template,
+                JournalFile **ret);
+
 void journal_file_close(JournalFile *j);
 
-int journal_file_open_reliably(const char *fname, int flags, mode_t mode, JournalFile *template, JournalFile **ret);
+int journal_file_open_reliably(
+                const char *fname,
+                int flags,
+                mode_t mode,
+                JournalMetrics *metrics,
+                JournalFile *template,
+                JournalFile **ret);
 
 int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret);
 
index 8f00aef..d8cfa66 100644 (file)
@@ -313,7 +313,7 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
                 journal_file_close(f);
         }
 
-        r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
+        r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, &s->system_metrics, s->system_journal, &f);
         free(p);
 
         if (r < 0)
@@ -2005,13 +2005,10 @@ static int system_journal_open(Server *s) {
                 if (!fn)
                         return -ENOMEM;
 
-                r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
+                r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, &s->system_metrics, NULL, &s->system_journal);
                 free(fn);
 
                 if (r >= 0) {
-                        journal_default_metrics(&s->system_metrics, s->system_journal->fd);
-
-                        s->system_journal->metrics = s->system_metrics;
                         s->system_journal->compress = s->compress;
 
                         server_fix_perms(s, s->system_journal, 0);
@@ -2037,7 +2034,7 @@ static int system_journal_open(Server *s) {
                          * if it already exists, so that we can flush
                          * it into the system journal */
 
-                        r = journal_file_open(fn, O_RDWR, 0640, NULL, &s->runtime_journal);
+                        r = journal_file_open(fn, O_RDWR, 0640, &s->runtime_metrics, NULL, &s->runtime_journal);
                         free(fn);
 
                         if (r < 0) {
@@ -2053,7 +2050,7 @@ static int system_journal_open(Server *s) {
                          * it if necessary. */
 
                         (void) mkdir_parents(fn, 0755);
-                        r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
+                        r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, &s->runtime_metrics, NULL, &s->runtime_journal);
                         free(fn);
 
                         if (r < 0) {
@@ -2063,9 +2060,6 @@ static int system_journal_open(Server *s) {
                 }
 
                 if (s->runtime_journal) {
-                        journal_default_metrics(&s->runtime_metrics, s->runtime_journal->fd);
-
-                        s->runtime_journal->metrics = s->runtime_metrics;
                         s->runtime_journal->compress = s->compress;
 
                         server_fix_perms(s, s->runtime_journal, 0);
index 77469e9..497f79c 100644 (file)
@@ -1116,7 +1116,7 @@ static int add_file(sd_journal *j, const char *prefix, const char *filename) {
                 return 0;
         }
 
-        r = journal_file_open(path, O_RDONLY, 0, NULL, &f);
+        r = journal_file_open(path, O_RDONLY, 0, NULL, NULL, &f);
         free(path);
 
         if (r < 0) {
index 313606f..627faa7 100644 (file)
@@ -79,9 +79,9 @@ int main(int argc, char *argv[]) {
         assert_se(mkdtemp(t));
         assert_se(chdir(t) >= 0);
 
-        assert_se(journal_file_open("one.journal", O_RDWR|O_CREAT, 0666, NULL, &one) == 0);
-        assert_se(journal_file_open("two.journal", O_RDWR|O_CREAT, 0666, NULL, &two) == 0);
-        assert_se(journal_file_open("three.journal", O_RDWR|O_CREAT, 0666, NULL, &three) == 0);
+        assert_se(journal_file_open("one.journal", O_RDWR|O_CREAT, 0666, NULL, NULL, &one) == 0);
+        assert_se(journal_file_open("two.journal", O_RDWR|O_CREAT, 0666, NULL, NULL, &two) == 0);
+        assert_se(journal_file_open("three.journal", O_RDWR|O_CREAT, 0666, NULL, NULL, &three) == 0);
 
         for (i = 0; i < N_ENTRIES; i++) {
                 char *p, *q;
index 9e1a4f5..7271a3f 100644 (file)
@@ -41,7 +41,7 @@ int main(int argc, char *argv[]) {
         assert_se(mkdtemp(t));
         assert_se(chdir(t) >= 0);
 
-        assert_se(journal_file_open("test.journal", O_RDWR|O_CREAT, 0666, NULL, &f) == 0);
+        assert_se(journal_file_open("test.journal", O_RDWR|O_CREAT, 0666, NULL, NULL, &f) == 0);
 
         dual_timestamp_get(&ts);