io.test revision 5821806d5e7f356e8fa4b058a389a808ea183019
1# 2007 August 21
2#
3# The author disclaims copyright to this source code.  In place of
4# a legal notice, here is a blessing:
5#
6#    May you do good and not evil.
7#    May you find forgiveness for yourself and forgive others.
8#    May you share freely, never taking more than you give.
9#
10#***********************************************************************
11#
12# The focus of this file is testing some specific characteristics of the 
13# IO traffic generated by SQLite (making sure SQLite is not writing out
14# more database pages than it has to, stuff like that).
15#
16
17set testdir [file dirname $argv0]
18source $testdir/tester.tcl
19
20db close
21sqlite3_simulate_device
22sqlite3 db test.db -vfs devsym
23
24# Test summary:
25#
26# io-1.* -  Test that quick-balance does not journal pages unnecessarily.
27#
28# io-2.* -  Test the "atomic-write optimization".
29#
30# io-3.* -  Test the IO traffic enhancements triggered when the 
31#           IOCAP_SEQUENTIAL device capability flag is set (no 
32#           fsync() calls on the journal file).
33#
34# io-4.* -  Test the IO traffic enhancements triggered when the 
35#           IOCAP_SAFE_APPEND device capability flag is set (fewer 
36#           fsync() calls on the journal file, no need to set nRec
37#           field in the single journal header).
38#
39# io-5.* -  Test that the default page size is selected and used 
40#           correctly.
41#           
42
43set ::nWrite 0
44proc nWrite {db} {
45  set bt [btree_from_db $db]
46  db_enter $db
47  array set stats [btree_pager_stats $bt]
48  db_leave $db
49  set res [expr $stats(write) - $::nWrite]
50  set ::nWrite $stats(write)
51  set res
52}
53
54set ::nSync 0
55proc nSync {} {
56  set res [expr {$::sqlite_sync_count - $::nSync}]
57  set ::nSync $::sqlite_sync_count
58  set res
59}
60
61do_test io-1.1 {
62  execsql {
63    PRAGMA auto_vacuum = OFF;
64    PRAGMA page_size = 1024;
65    CREATE TABLE abc(a,b);
66  }
67  nWrite db
68} {2}
69
70# Insert into the table 4 records of aproximately 240 bytes each.
71# This should completely fill the root-page of the table. Each
72# INSERT causes 2 db pages to be written - the root-page of "abc"
73# and page 1 (db change-counter page).
74do_test io-1.2 {
75  set ret [list]
76  execsql { INSERT INTO abc VALUES(1,randstr(230,230)); }
77  lappend ret [nWrite db]
78  execsql { INSERT INTO abc VALUES(2,randstr(230,230)); }
79  lappend ret [nWrite db]
80  execsql { INSERT INTO abc VALUES(3,randstr(230,230)); }
81  lappend ret [nWrite db]
82  execsql { INSERT INTO abc VALUES(4,randstr(230,230)); }
83  lappend ret [nWrite db]
84} {2 2 2 2}
85
86# Insert another 240 byte record. This causes two leaf pages
87# to be added to the root page of abc. 4 pages in total
88# are written to the db file - the two leaf pages, the root
89# of abc and the change-counter page.
90do_test io-1.3 {
91  execsql { INSERT INTO abc VALUES(5,randstr(230,230)); }
92  nWrite db
93} {4}
94
95# Insert another 3 240 byte records. After this, the tree consists of 
96# the root-node, which is close to empty, and two leaf pages, both of 
97# which are full. 
98do_test io-1.4 {
99  set ret [list]
100  execsql { INSERT INTO abc VALUES(6,randstr(230,230)); }
101  lappend ret [nWrite db]
102  execsql { INSERT INTO abc VALUES(7,randstr(230,230)); }
103  lappend ret [nWrite db]
104  execsql { INSERT INTO abc VALUES(8,randstr(230,230)); }
105  lappend ret [nWrite db]
106} {2 2 2}
107
108# This insert should use the quick-balance trick to add a third leaf
109# to the b-tree used to store table abc. It should only be necessary to
110# write to 3 pages to do this: the change-counter, the root-page and
111# the new leaf page.
112do_test io-1.5 {
113  execsql { INSERT INTO abc VALUES(9,randstr(230,230)); }
114  nWrite db
115} {3}
116
117ifcapable atomicwrite {
118
119#----------------------------------------------------------------------
120# Test cases io-2.* test the atomic-write optimization.
121#
122do_test io-2.1 {
123  execsql { DELETE FROM abc; VACUUM; }
124} {}
125
126# Clear the write and sync counts.
127nWrite db ; nSync
128
129# The following INSERT updates 2 pages and requires 4 calls to fsync():
130#
131#   1) The directory in which the journal file is created,
132#   2) The journal file (to sync the page data),
133#   3) The journal file (to sync the journal file header),
134#   4) The database file.
135#
136do_test io-2.2 {
137  execsql { INSERT INTO abc VALUES(1, 2) }
138  list [nWrite db] [nSync]
139} {2 4}
140
141# Set the device-characteristic mask to include the SQLITE_IOCAP_ATOMIC,
142# then do another INSERT similar to the one in io-2.2. This should
143# only write 1 page and require a single fsync().
144# 
145# The single fsync() is the database file. Only one page is reported as
146# written because page 1 - the change-counter page - is written using
147# an out-of-band method that bypasses the write counter.
148#
149sqlite3_simulate_device -char atomic
150do_test io-2.3 {
151  execsql { INSERT INTO abc VALUES(3, 4) }
152  list [nWrite db] [nSync]
153} {1 1}
154
155# Test that the journal file is not created and the change-counter is
156# updated when the atomic-write optimization is used.
157#
158do_test io-2.4.1 {
159  execsql {
160    BEGIN;
161    INSERT INTO abc VALUES(5, 6);
162  }
163  sqlite3 db2 test.db -vfs devsym
164  execsql { SELECT * FROM abc } db2
165} {1 2 3 4}
166do_test io-2.4.2 {
167  file exists test.db-journal
168} {0}
169do_test io-2.4.3 {
170  execsql { COMMIT }
171  execsql { SELECT * FROM abc } db2
172} {1 2 3 4 5 6}
173db2 close
174
175# Test that the journal file is created and sync()d if the transaction
176# modifies more than one database page, even if the IOCAP_ATOMIC flag
177# is set.
178#
179do_test io-2.5.1 {
180  execsql { CREATE TABLE def(d, e) }
181  nWrite db ; nSync
182  execsql {
183    BEGIN;
184    INSERT INTO abc VALUES(7, 8);
185  }
186  file exists test.db-journal
187} {0}
188do_test io-2.5.2 {
189  execsql { INSERT INTO def VALUES('a', 'b'); }
190  file exists test.db-journal
191} {1}
192do_test io-2.5.3 {
193  execsql { COMMIT }
194  list [nWrite db] [nSync]
195} {3 4}
196
197# Test that the journal file is created and sync()d if the transaction
198# modifies a single database page and also appends a page to the file.
199# Internally, this case is handled differently to the one above. The
200# journal file is not actually created until the 'COMMIT' statement
201# is executed.
202#
203# Changed 2010-03-27:  The size of the database is now stored in 
204# bytes 28..31 and so when a page is added to the database, page 1
205# is immediately modified and the journal file immediately comes into
206# existance.  To fix this test, the BEGIN is changed into a a
207# BEGIN IMMEDIATE and the INSERT is omitted.
208#
209do_test io-2.6.1 {
210  execsql {
211    BEGIN IMMEDIATE;
212    -- INSERT INTO abc VALUES(9, randstr(1000,1000));
213  }
214  file exists test.db-journal
215} {0}
216do_test io-2.6.2 {
217  # Create a file at "test.db-journal". This will prevent SQLite from
218  # opening the journal for exclusive access. As a result, the COMMIT
219  # should fail with SQLITE_CANTOPEN and the transaction rolled back.
220  #
221  file mkdir test.db-journal
222  catchsql {
223    INSERT INTO abc VALUES(9, randstr(1000,1000));
224    COMMIT
225  }
226} {1 {unable to open database file}}
227do_test io-2.6.3 {
228  file delete -force test.db-journal
229  catchsql { COMMIT }
230} {0 {}}
231do_test io-2.6.4 {
232  execsql { SELECT * FROM abc }
233} {1 2 3 4 5 6 7 8}
234
235# Test that if the database modification is part of multi-file commit,
236# the journal file is always created. In this case, the journal file
237# is created during execution of the COMMIT statement, so we have to
238# use the same technique to check that it is created as in the above 
239# block.
240file delete -force test2.db test2.db-journal
241ifcapable attach {
242  do_test io-2.7.1 {
243    execsql {
244      ATTACH 'test2.db' AS aux;
245      PRAGMA aux.page_size = 1024;
246      CREATE TABLE aux.abc2(a, b);
247      BEGIN;
248      INSERT INTO abc VALUES(9, 10);
249    }
250    file exists test.db-journal
251  } {0}
252  do_test io-2.7.2 {
253    execsql { INSERT INTO abc2 SELECT * FROM abc }
254    file exists test2.db-journal
255  } {0}
256  do_test io-2.7.3 {
257    execsql { SELECT * FROM abc UNION ALL SELECT * FROM abc2 }
258  } {1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10}
259  do_test io-2.7.4 {
260    file mkdir test2.db-journal
261    catchsql { COMMIT }
262  } {1 {unable to open database file}}
263  do_test io-2.7.5 {
264    file delete -force test2.db-journal
265    catchsql { COMMIT }
266  } {1 {cannot commit - no transaction is active}}
267  do_test io-2.7.6 {
268    execsql { SELECT * FROM abc UNION ALL SELECT * FROM abc2 }
269  } {1 2 3 4 5 6 7 8}
270}
271
272# Try an explicit ROLLBACK before the journal file is created.
273#
274do_test io-2.8.1 {
275  execsql {
276    BEGIN;
277    DELETE FROM abc;
278  }
279  file exists test.db-journal
280} {0}
281do_test io-2.8.2 {
282  execsql { SELECT * FROM abc }
283} {}
284do_test io-2.8.3 {
285  execsql {
286    ROLLBACK;
287    SELECT * FROM abc;
288  }
289} {1 2 3 4 5 6 7 8}
290
291# Test that the atomic write optimisation is not enabled if the sector
292# size is larger than the page-size.
293#
294do_test io-2.9.1 {
295  db close
296  sqlite3 db test.db
297  sqlite3_simulate_device -char atomic -sectorsize 2048
298  execsql {
299    BEGIN;
300    INSERT INTO abc VALUES(9, 10);
301  }
302  file exists test.db-journal
303} {1}
304do_test io-2.9.2 {
305  execsql { ROLLBACK; }
306  db close
307  file delete -force test.db test.db-journal
308  sqlite3 db test.db -vfs devsym
309  execsql {
310    PRAGMA auto_vacuum = OFF;
311    PRAGMA page_size = 2048;
312    CREATE TABLE abc(a, b);
313  }
314  execsql {
315    BEGIN;
316    INSERT INTO abc VALUES(9, 10);
317  }
318  file exists test.db-journal
319} {0}
320do_test io-2.9.3 {
321  execsql { COMMIT }
322} {}
323
324# Test a couple of the more specific IOCAP_ATOMIC flags 
325# (i.e IOCAP_ATOMIC2K etc.).
326#
327do_test io-2.10.1 {
328  sqlite3_simulate_device -char atomic1k
329  execsql {
330    BEGIN;
331    INSERT INTO abc VALUES(11, 12);
332  }
333  file exists test.db-journal
334} {1}
335do_test io-2.10.2 {
336  execsql { ROLLBACK }
337  sqlite3_simulate_device -char atomic2k
338  execsql {
339    BEGIN;
340    INSERT INTO abc VALUES(11, 12);
341  }
342  file exists test.db-journal
343} {0}
344do_test io-2.10.3 {
345  execsql { ROLLBACK }
346} {}
347
348do_test io-2.11.0 {
349  execsql { 
350    PRAGMA locking_mode = exclusive;
351    PRAGMA locking_mode;
352  }
353} {exclusive exclusive}
354do_test io-2.11.1 {
355  execsql { 
356    INSERT INTO abc VALUES(11, 12);
357  }
358  file exists test.db-journal
359} {0}
360
361do_test io-2.11.2 {
362  execsql { 
363    PRAGMA locking_mode = normal;
364    INSERT INTO abc VALUES(13, 14);
365  }
366  file exists test.db-journal
367} {0}
368
369} ;# /* ifcapable atomicwrite */
370
371#----------------------------------------------------------------------
372# Test cases io-3.* test the IOCAP_SEQUENTIAL optimization.
373#
374sqlite3_simulate_device -char sequential -sectorsize 0
375ifcapable pager_pragmas {
376  do_test io-3.1 {
377    db close
378    file delete -force test.db test.db-journal
379    sqlite3 db test.db -vfs devsym
380    db eval {
381      PRAGMA auto_vacuum=OFF;
382    }
383    # File size might be 1 due to the hack to work around ticket #3260.
384    # Search for #3260 in os_unix.c for additional information.
385    expr {[file size test.db]>1}
386  } {0}
387  do_test io-3.2 {
388    execsql { CREATE TABLE abc(a, b) }
389    nSync
390    execsql {
391      PRAGMA temp_store = memory;
392      PRAGMA cache_size = 10;
393      BEGIN;
394      INSERT INTO abc VALUES('hello', 'world');
395      INSERT INTO abc SELECT * FROM abc;
396      INSERT INTO abc SELECT * FROM abc;
397      INSERT INTO abc SELECT * FROM abc;
398      INSERT INTO abc SELECT * FROM abc;
399      INSERT INTO abc SELECT * FROM abc;
400      INSERT INTO abc SELECT * FROM abc;
401      INSERT INTO abc SELECT * FROM abc;
402      INSERT INTO abc SELECT * FROM abc;
403      INSERT INTO abc SELECT * FROM abc;
404      INSERT INTO abc SELECT * FROM abc;
405      INSERT INTO abc SELECT * FROM abc;
406    }
407    # File has grown - showing there was a cache-spill - but there 
408    # have been no calls to fsync(). The file is probably about 30KB.
409    # But some VFS implementations (symbian) buffer writes so the actual
410    # size may be a little less than that. So this test case just tests
411    # that the file is now greater than 20000 bytes in size.
412    list [expr [file size test.db]>20000] [nSync]
413  } {1 0}
414  do_test io-3.3 {
415    # The COMMIT requires a single fsync() - to the database file.
416    execsql { COMMIT }
417    list [file size test.db] [nSync]
418  } {39936 1}
419}
420
421#----------------------------------------------------------------------
422# Test cases io-4.* test the IOCAP_SAFE_APPEND optimization.
423#
424sqlite3_simulate_device -char safe_append
425
426# With the SAFE_APPEND flag set, simple transactions require 3, rather
427# than 4, calls to fsync(). The fsync() calls are on:
428#
429#   1) The directory in which the journal file is created, (unix only)
430#   2) The journal file (to sync the page data),
431#   3) The database file.
432#
433# Normally, when the SAFE_APPEND flag is not set, there is another fsync()
434# on the journal file between steps (2) and (3) above.
435#
436set expected_sync_count 2
437if {$::tcl_platform(platform)=="unix"} {
438  ifcapable dirsync {
439    incr expected_sync_count
440  }
441}
442
443do_test io-4.1 {
444  execsql { DELETE FROM abc }
445  nSync
446  execsql { INSERT INTO abc VALUES('a', 'b') }
447  nSync
448} $expected_sync_count
449
450# With SAFE_APPEND set, the nRec field of the journal file header should
451# be set to 0xFFFFFFFF before the first journal sync. The nRec field
452# occupies bytes 8-11 of the journal file.
453#
454do_test io-4.2.1 {
455  execsql { BEGIN }
456  execsql { INSERT INTO abc VALUES('c', 'd') }
457  file exists test.db-journal
458} {1}
459if {$::tcl_platform(platform)=="unix"} {
460  do_test io-4.2.2 {
461    hexio_read test.db-journal 8 4
462  } {FFFFFFFF}
463}
464do_test io-4.2.3 {
465  execsql { COMMIT }
466  nSync
467} $expected_sync_count
468sqlite3_simulate_device -char safe_append
469
470# With SAFE_APPEND set, there should only ever be one journal-header
471# written to the database, even though the sync-mode is "full".
472#
473do_test io-4.3.1 {
474  execsql {
475    INSERT INTO abc SELECT * FROM abc;
476    INSERT INTO abc SELECT * FROM abc;
477    INSERT INTO abc SELECT * FROM abc;
478    INSERT INTO abc SELECT * FROM abc;
479    INSERT INTO abc SELECT * FROM abc;
480    INSERT INTO abc SELECT * FROM abc;
481    INSERT INTO abc SELECT * FROM abc;
482    INSERT INTO abc SELECT * FROM abc;
483    INSERT INTO abc SELECT * FROM abc;
484    INSERT INTO abc SELECT * FROM abc;
485    INSERT INTO abc SELECT * FROM abc;
486  }
487  expr {[file size test.db]/1024}
488} {43}
489ifcapable pager_pragmas {
490  do_test io-4.3.2 {
491    execsql {
492      PRAGMA synchronous = full;
493      PRAGMA cache_size = 10;
494      PRAGMA synchronous;
495    }
496  } {2}
497}
498do_test io-4.3.3 {
499  execsql {
500    BEGIN;
501    UPDATE abc SET a = 'x';
502  }
503  file exists test.db-journal
504} {1}
505if {$tcl_platform(platform) != "symbian"} {
506  # This test is not run on symbian because the file-buffer makes it
507  # difficult to predict the exact size of the file as reported by 
508  # [file size].
509  do_test io-4.3.4 {
510    # The UPDATE statement in the statement above modifies 41 pages 
511    # (all pages in the database except page 1 and the root page of 
512    # abc). Because the cache_size is set to 10, this must have required
513    # at least 4 cache-spills. If there were no journal headers written
514    # to the journal file after the cache-spill, then the size of the
515    # journal file is give by:
516    #
517    #    <jrnl file size> = <jrnl header size> + nPage * (<page-size> + 8)
518    #
519    # If the journal file contains additional headers, this formula
520    # will not predict the size of the journal file.
521    #
522    file size test.db-journal
523  } [expr 512 + (1024+8)*41]
524}
525
526#----------------------------------------------------------------------
527# Test cases io-5.* test that the default page size is selected and
528# used correctly.
529#
530set tn 0
531foreach {char                 sectorsize pgsize} {
532         {}                     512      1024
533         {}                    1024      1024
534         {}                    2048      2048
535         {}                    8192      8192
536         {}                   16384      8192
537         {atomic}               512      8192
538         {atomic512}            512      1024
539         {atomic2K}             512      2048
540         {atomic2K}            4096      4096
541         {atomic2K atomic}      512      8192
542         {atomic64K}            512      1024
543} {
544  incr tn
545  if {$pgsize>$::SQLITE_MAX_PAGE_SIZE} continue
546  db close
547  file delete -force test.db test.db-journal
548  sqlite3_simulate_device -char $char -sectorsize $sectorsize
549  sqlite3 db test.db -vfs devsym
550  db eval {
551    PRAGMA auto_vacuum=OFF;
552  }
553  ifcapable !atomicwrite {
554    if {[regexp {^atomic} $char]} continue
555  }
556  do_test io-5.$tn {
557    execsql {
558      CREATE TABLE abc(a, b, c);
559    }
560    expr {[file size test.db]/2}
561  } $pgsize
562}
563
564sqlite3_simulate_device -char {} -sectorsize 0
565finish_test
566