We're having a problem on a wiki site (using mysql 5.5.9). There is a text
table, and a revision
table where revision.rev_text_id
is a foreign key to text.old_id
(there isn't a new one, ask wikimedia).
There is an INSERT INTO text...
, then INSERT INTO revision...
which uses the new old_id
/rev_text_id
received from the previous query.
The second query works, the first - I'm not sure. The thing is after the whole thing is gone through (with a few dozen more queries) the revision
row is there, with a fresh new value in it's rev_text_id
column.
However, the text
row isn't there. Funny thing - the text
table auto increment advances, as the next action skips a value for it's old_id
, the missing value matching what we have in revision
table.
Is there a chance the first query makes the auto increment index advance though the row isn't actually inserted? What causes this?
ADDTITION
When I copied the INSERT INTO text
query from the logs and ran it on the server - it executed fine (row was added to table)
EDIT
Full queries are:
INSERT INTO text (old_id,old_text,old_flags) VALUES (NULL,'{text input by user}','utf8');
INSERT INTO `revision` (rev_id,rev_page,rev_text_id,rev_comment,rev_minor_edit,rev_user,rev_user_text,rev_timestamp,rev_deleted,rev_len,rev_parent_id,rev_sha1) VALUES (NULL,'{pageId}','{textId}','{comment}','{isMinor}','{userId}','{userName}','{TS}','{isDeleted}','{length}','{parentRevision}','{HASH}')"
Also, if relevant, text
table is running InnoDB and revision
is MyISAM.
EDIT
More information from logs:
Bad request
BEGIN
INSERT INTO `text`
INSERT INTO `revision`
UPDATE `page` SET page_latest
INSERT INTO `recentchanges`
INSERT INTO `cu_changes`
SELECT wl_user FROM `watchlist`
SELECT user_id FROM `user`
SELECT user_id,user_name,user_real_name,user_password,user_newpassword,user_newpass_time,user_email,user_touched,use
SELECT ug_group FROM `user_groups`
SELECT up_property,up_value FROM `user_properties`
SELECT user_id,user_name,user_real_name,user_password,user_newpassword,user_newpass_time,user_email,user_touched,use
SELECT ug_group FROM `user_groups`
SELECT up_property,up_value FROM `user_properties`
SELECT lc_value FROM `l10n_cache`
SELECT lc_value FROM `l10n_cache`
...
A few more SELECT
s. Pause of activity for 2 sec, then the log has commands from a new user, no more queries from this user (on same thread).
An edit by same user to a different page, which went well:
BEGIN
INSERT INTO `text`
INSERT INTO `revision`
UPDATE `page` SET page_latest
INSERT INTO `recentchanges`
INSERT INTO `cu_changes`
SELECT wl_user FROM `watchlist`
COMMIT
BEGIN
UPDATE `watchlist` SET wl_notificationtimestamp
COMMIT
BEGIN
SELECT user_id FROM `user`
SELECT user_id,user_name,user_real_name,user_password,user_newpasswo
SELECT ug_group FROM `user_groups`
SELECT up_property,up_value FROM `user_properties`
SELECT * FROM `user`
SELECT up_property,up_value FROM `user_properties`
INSERT INTO `logging`
UPDATE `user` SET user_editcount=user_editcount+1
SELECT 1 FROM `user`
UPDATE `user` SET user_touched = '20121227211743'
COMMIT
etc.
Is there a chance the first query makes the auto increment index advance though the row isn't actually inserted? What causes this?
Yes, when you try to insert a row in a table with an auto_increment column, the insertion will fail, but the auto_increment value will be incremented by 1.
If you decide to convert the tables to InnoDB, I suggest you to use transactions. This way if one the queries fails none of them will be inserted.