A Different Set of Stories

A humming bird flew into the house last night when we opened the front door to head out for dinner. It immediately flew up to the skylight. No chance to get it. So we turn out all the lights, left the back sliding door open and hoped for the best. When we got back from dinner, we discovered our plan almost worked. When it got dark, the bird abandoned the skylight, but went for the high window over the front door. The overly bright street light shines in that window. We tried various things that had no chance of working and gave up. When we went into our room and turned on the light, the bird flew in. This is good. We have low ceilings. And the poor thing was exhausted. After a couple of tries, I managed to scoop it up in a towel. I left it outside on the towel near water and blooming flowers. I pretty much that it was going to die of exhaustion and fear. Apparently not. This morning I found the towel just like I left it and the bird gone. Since the towel was not disturbed, I'm figuring the bird is off telling the story of getting away.

This morning, I dragged myself out of bed in the dark (I had switching to and from DST) and got myself to work nice and early. Locked out of the building. It is a card entry building and nobodies' cards worked. Thought it might be because of DST, so I waited the 20 minutes 'til the building thought it was time to unlock the doors. Nope. Waited another 30 minutes until "management" showed up and did some hocus-pocus to open the doors. Not a good start for the day.

This morning, a client using FireFox 27, submitted a form at our website. What we got was each form element three times! I'm not talking about the form being submitted multiple times, but each element being bundled up three times within a single POST operation.

I've upped my logging to watch for this to make sure I'm really getting the data three times and it isn't something I'm doing. I can't see how it is something I'm doing, though.

It didn't really affect much. I sanitize pretty carefully. We only noticed because a couple of the elements allow multiple values and the same value showed up three times rather than three different values.

This one is new to me!

There is a philosophy adopted my many open source projects and, probably, all the big ones that it is good to work on as many platforms as possible. I generally agree with this.

I don't agree with it when the project must dumb down to make sure it works everywhere. Like when it must compromise by not using native features on one platform because it doesn't exist on another. It is a problem when you want to work on OSX, MS whatever, and Linux.

But a bigger problem in my world is when a project wants to be use many different storage systems. For instance, if a project is going to use an sql database, and it won't pick ONE, it must be no smarter than the dumbest one in the offering and that is usually pretty damn dumb!

I am getting ready to replace the production email server at work with a new one using a different set of components. Right now, I handle virtual domains and virtual users in a hard-to-maintain fashion using courier-auth and courier-imapd and some hocus-pocus involving flat files.

I want to put as much as possible in postgresql. To that end I'm using DBmail with a few of my own database extensions.

I have a test server running. And while troubleshooting a delivery problem I turned up verbosity on all bits of postfix. And a warning appeared in the logs. A warning on a piece I thought was settled long ago. It seems postfix thought the localhost domain was in mydestinations and virtual_mailbox_domains. Well, I knew for a fact (and double checked) that localhost was not in the virtual tables anywhere.

When postfix is looking to see if we accept mail for a given domain, it checks a bunch of places. One of the last is the virtual_mailbox_domains list. Postfix is not asking for a yes/no response from the database; rather, it is counting the NUMBER OF ROWS returned by the database query which will be 0 or 1. I'm sure there must be a good reason for this. I suspect it allows the same code to be used to support a lot of ways of storing this information and some can't return boolean.

I wrote a series of functions in postgresql so that the postfix queries call my functions. And that is the rub. If you have a function foo(domain text) that looks like this:

CREATE FUNCTION foo (domain text) RETURNS text AS $$ SELECT domain FROM allowed_domains WHERE domain = $1 $$ LANGUAGE sql

And you call this function like this:

SELECT foo('example.com');

You will always get one row returned and it doesn't matter if example.com is in the database or not. That row will either be "example.com" or "NULL". foo() returns no rows or 1 row, but the select that calls it returns the result of foo or null. Postfix counts rows and it doesn't care that the value of the row is unassigned.

It has also been suggested that returning arbitrary text might be an issue for postfix. I have seen no evidence of that, but returning a "1" when there is a match seems safer. That means that I could do something like this:

SELECT 1 FROM (select domain from foo('%s') where domain is not null) x

Postfix substitutes the necessary key value for the '%s'. This call to my function returns one row with a value of 1 if the domain is found and returns no rows if the domain is not found.

But this is a bit clunky looking. The answer is a view.

CREATE VIEW allowed_domain AS
SELECT domain FROM allowed_domains
WHERE domain <> 'specialname' AND
domain <> 'otherspecialname';

Then the postfix query can look like this:

SELECT 1 FROM allowed_domain WHERE domain = '%s';

Better. Much, much better. It is an error to return more than one row to postfix. The table allowed_domains is not allowed to have duplicate entries so we will "never" return more than 1 row and now we can return 0 rows.

Last Sunday night after 8:30 pm, our offices were broken into. A small window in my office was smashed in. My monitor and my laptop was taken. Petty cash was taken. A small travel case with a cheap digital camera was taken. The feeling is this: it was the local druggie youth in our area. The druggie youth we know thinks this, too.

The good news is that we recently installed cameras and a DVR. The bad news is that I didn't set it up quite right and it didn't record anything. sigh.

It also seems that our suite has been hit in the past. Wish someone had told us we had some reasonable likelihood of theft! Well, we are keeping our camera system and installing a monitored alarm system.

I hate being robbed.

I ordered two new computers for work for the firewall/proxy machines. I had two requirements: 1) Run Linux; and, 2) have three Ethernet ports. I picked them up today. To get the three nics, I had to not get the cheapest motherboards. Fine. More horsepower than I need for the work the machines do. Oh well.

Picked them up this afternoon. As I expected, these motherboards do not have ps2 connectors. So they aren't going to get connected to my ps2 KVM switch. Bother. Not the end of the world, but a bother.

As I didn't expect at all, it doesn't have a vga connector, either. REAL BOTHER. It has hdmi, and dvi and something called "display port" with a connector I don't recognize. I know that vga is being phased out. I don't even mind. But it does mean that I am now in the very slow transition period in which I will have machines that aren't easily re-purposed, can't be hooked up to my ps2/vga KVM in the server room, and will be a pain in the ass to maintain.

I just happen to have a monitor (the one I was using before I needed double size) that will work with the new machines. I just don't have cables for it. Guess I need to go to the store.

I wrote a small debugging library to use in the business logic at work. It works very, very well but the log files are sometimes a bit cryptic because multiple things are going on at the same time which means the debugging logs have entries from different processes intertwined.

This morning, I finally looked at that library again. Duh. Every process has its own pid, Lynn. Just prefix each log entry with the time stamp and the pid. Done.

Now when I identify the pid of the process I want to follow, I grep the log for it. It provides a beautiful list of just the info I want. Much, much better. I can grep the log for specific user data, for instance, which gives me the pid, then grep the pid. I think I'll write a little script that does the second part for me. Simple enough.

We have had cbeyond as our internet/phone provider for five years. Until a couple of weeks ago, they gave us a pri circuit that fed our own aging pbx. I have been looking for a reason to dump it. It is old, I have very limited control over it and it is a disaster waiting to happen.

I was thinking I would set up buy the necessary cards from digium and run asterisk locally. Just get cbeyond to give me a sip line. They made us a deal for increased internet bandwidth ( from 2 to 3 T-1) plus a hosted asterisk pbx that I would have a great deal of control over. All for $300 a month.

We talked it over and decided to go for it.

We have been on the hosted pbx for a week. It is a fucking disaster. It is asterisk with freepbx on top. We bought the phones they suggested and told them what the initial configuration should be. We changed as part our move to the new office. It took twice as long as the longest time they said and the phones did not work as we asked.

I have been messing with the damn web config tool for a week. And it still does not perform correctly. I have been on the phone with tech support, but they don't know crap about this pbx. Cbeyond bought aretta but their support people are groping along with me for solutions.

One basic routing is to change the call routing when we open in the morning and change it back when we close. I set up a test with a number we don't use and it worked fine. The real numbers however are flaky as hell. For the last two days, it put us into night mode at 2pm instead of 5pm. I think it's a timezone issue. The damn system comes with extensible "feature codes" that allow you to type "*something" to affect the phone status. Logging into or out of the ACD queue, for instance. Doesn't work as advertized. Further, there is no indication whether a phone is logged in or not. What a fucking joke.

I was on the phone for 90 minutes with tier 2 and 3 tech support yesterday. They messed with this and that, remotely reprogrammed one of the phones, found a few work arounds for some issues. And today, it is all crap again.

Well. the old system is sitting in the back room. I think I'm going to hang it on the wall and tell cbeyond what they can do with their hosted pbx.

it might be time to look for a new provider for everything. They have fumbled the ball on every possession in the transition to the new phones and in the move from the old office to here.

And they have changed their trouble ticket system. You log into their sight and get a goddamn interactive flash presentation of your trouble tickets. I can't fucking read the tiny ass print they decided to use and I can't re-size it. What kind of fucking anti-ADA is that! (Trustwave has this same issue.)

Really, guys. Really?

Well, hell. That's the end of my trying to use archiveopteryx. I've spent a week working out the kinks in the configuration of postfix and archiveopteryx only to find that they don't play with each other all that nicely. I really like the idea of storing email in postgresql which what archiveopteryx is good at.

But to do that, I have to tell postfix to "relay" mail to it over either smtp or lmtp. And all that is fine, too. Except that if I set the mailbox_transport configuration to do that, postfix immediately stops rejecting unlisted recipients. As far as postfix is concerned it is just a mail relay. Okay, I can live with that, too.

But archiveopteryx does not doesn't reject, it bounces, mail for unknown recipients. And that I can't live with. I want mail to unknowns to simple fall into the bit bucket. I am not interested in becoming just another great backsplatter server for spammers to use.

I suppose I could try and modify archiveopteryx source. Hell, I'll go have a look anyway. But, damn that is the hard way.

As I have written more and more databases within a given PostgreSQL cluster, I have run into a bit of a problem with "version control" of sets of utility domains, types, functions and the like.

Every database created comes with a default schema named public. And anything created in that database is automatically part of the public schema. Although somethings are private to a given database, somethings are not.

Domains and functions in the public schema are available to other databases. If I redefine a function in my new database that has been used in other databases, I've just created a problem in the other databases. Not good.

I've decided, therefore, to create a schema for each database and create everything inside that schema. Then I can change the schema search path to check for the local schema FIRST, then public. Or not check public at all.

I think this works. I'm going to try it for a bit and see if it deals with the issue.

EDIT. It seems that the default search_path starts with "$user" which means that the default schema for everything has the same name as the current user (assuming the schema has been created). I'll just create it that way and make sure that I always connect to a particular database as the user that owns the database. Problem solved.