Tyrone Slothrop wrote:
<cut to the chase>

Right. Can you CONCLUSIVELY say that the corruption you get does NOT
show up as a malformed SQL query logged by this script?


STOP! you don't know the bad query is coming from this area at all.

How many other process communicate with mysqld? how many COULD
communicate with it, in theory?

I mean is there an old cron script or a test script somewhere still
running..

Years ago a story was going round a big software house in Johannesburg.
About te code that always crashed about every year, requiring its prime
author to make a trip to Durban to fix it.

Several year went by before it was noted that the author, a keen racing
fan, always seemed to be 'away in Durban' fixing this code during the
week the Durban Races were on...



That is almost definitely an update table ...set fields with no where..



It is always a certain set of fields which are
Next question: is the field set that is altered, *every* altered set
that is, possible (and consistent) with the ajax partial update program?

If not something else is.maybe doing it.


You should - and its a vile thought - turn mysqld logging on.

Generally this is helpful when you can reproduce the fault easily. Its
bad news to leave running for a year to trap the odd 'neutrino' SQL
statement ;-)



Since the PHP script called
And, I take it, that this log is consistent and correct in every way?

which means that either you've missed some code that does a rogue
update, or some other process is doing it..

Axel Schwenke wrote:
Can you shed more light on the mysql-proxy?

I had not heard of that..

It looks a most useful tool.

On 2 July, 09:57, The Natural Philosopher <t... (AT) invalid (DOT) invalid>
wrote:
I'm sorry, the temptation is too great!

http://lmgtfy.com/?q=mysql-proxy

where you will find:
http://dev.mysql.com/tech-resources/...ngstarted.html

The Natural Philosopher <tnp (AT) invalid (DOT) invalid> wrote:
http://forge.mysql.com/wiki/MySQL_Proxy

Written by a MySQL^WSun^WOracle employee. All the time critical part
is C but you can plug Lua functions at certain stages of request and
response handling.

Some use cases: load balancing, read-write splitting + load balancing
(to distribute connections on a replication cluster). Statistical
analysis. I.e. proxy is used to collect data for the Enterprise Monitor
http://www.mysql.com/products/enterprise/monitor.html


XL

In article <i0k9de$54p$1 (AT) news (DOT) albasani.net>, The Natural Philosopher <tnp (AT) invalid (DOT) invalid> wrote:
More to the point, can he conclusively state that all queries that update the
database are being logged by this script? Of course not. There is no assurance
that the faulty query has any connection at all with the script that's doing
the logging.
[...]
Or with a WHERE clause that matches all rows.
[...]
That is the only way to be sure that every query gets logged. Right now, the
only queries being logged are those that the application has been told to log.
[...]
The OP doesn't seem to have considered that possibility...

Captain Paralytic wrote:

It was more 'how in general did you find it? what problems did it solve
for YOU? etc. etc. info I was after..

Axel Schwenke wrote:

Looks like a useful tool if I ever want it.

On Fri, 02 Jul 2010 00:43:23 -0500, Gordon Burditt wrote:
This. And validate the input anyway. And check it again. If your delete
operation is supposed to delete one record, run a select on that
criteria first *every time*, spit back an error, log it, and *don't run
the query*, if the number of records is anything OTHER than exactly what
you're expecting: 1.

--
40. I will be neither chivalrous nor sporting. If I have an unstoppable
superweapon, I will use it as early and as often as possible instead
of keeping it in reserve.
--Peter Anspach's list of things to do as an Evil Overlord

On Fri, 02 Jul 2010 12:55:03 GMT, "Peter H. Coffin"
<hellsop (AT) ninehells (DOT) com> wrote:

Thank you all for your opinions and suggestions.

I truly loathe to turn on MySQL logging since this is a production
server with sites which provide the income for my client (who is about
as anal-compulsive person as I have ever known). Unfortunately, I
cannot sandbox the application and place it under the same stress as
in the production environment. I am strongly encouraging the client
to lease a dedicated MySQL server. As load has been increasing, I
have moved the nameserver service from the server and in the process
of moving all mail services.

The AJAX code uses post and SSL, and passes raw data directly from the
form. The real work is done in PHP. I much prefer using 100%
server-side validation but the client had to have this feature and
though I fought against it, in the end, we all do what the client
wants since she/he is the one paying my bills.

Yes, it is very possible that an unescaped binary character is causing
the bad query -- at least that is my present theory. I am testing a
revised script which will be put in production in the next day or two.
The downside is that it will take months to know if the fix worked.

BTW, there is no direct access to the form by Google. The only access
to the form is via another form or by providing a unique query string.
If there is no match on the hash or the form has been completed, the
user is redirected to another page. The query string is used only to
identify the record and, once that record is called, is never used in
any way to provide the basis of an update query.

I will follow up on a couple suggestions in this thread.

Again, thank you all.