Proposal - simple/cheap failover strategy

[lvzardoz]

One thing I think most web sites dread is when something goes wrong (DNS down, server down, network down, etc.) and the visitor sees a message such as "we can't find XYZ.com").

Major sites use a "failover" system where all web documents are duplicated/mirrored at another site. An external "site-checker" determines if the primary site is up and if not it redirects the user to the secondary or tertiary site.

Zoneedit.com is a free (for the first 5) Managed DNS Service. For 1 "zone credit" ($5 to $11 per year) they will also provide a "failover service" to check your web site and redirect as necessary.

Now, I don't need full web site mirroring. I just don't want customers to think my web site doesn't exist if there happens to be some downtime. When my site is down for whatever reason I would just like to see a simple message saying that Sorry, the web site is down temporarily and the visitor should try again in a few minutes.

This should use minimal disk and bandwidth resources.

If JAGUARPC would make an agreement with another reliable HOST (hopefully geographically distant and on a different network/backbone) to make a standard mini-failover web site accessible by IP or URL, then we could point to the other site URL containing the Sorry Message as our failover "failure URL".

Preferably the message could also be customized by each user in this manner:

failure URL: http://sorry.otherhost.com/?Here%20is%20my%Message

So, what do you folks think? Any alternate suggestions?

Bob Carroll
Vegas Beltway Digital Services

From Zoneedit's FAQ:

How does your failover service work?

Failover monitoring service works when you have two or more web servers running the same (or similar) web site.

First, you need the IP address for your domain, and get it working - then you add the failover monitor. The failover monitor watches your web server(s) by hitting a URL you specify and looking for text in the results.

When the system detects that one of them is having an error, and the others aren't, it pulls the IP address out of the list. If none of the IP's are responding and you have a failure IP/URL defined, then it points the site to the failure IP/URL. If the system can't get to all of the IP's then it assumes that the fault is it's own connection - and it takes no action.

If any of your IP's come back online they are restored. This effectively and safely keeps your site online - even if one of your web servers is down.

The average failure detection time is 5 minutes. This time varies depending on the speed of your site and the nature of the failure. Recovery times are faster, averaging 3 minutes.

[mattsiegman]

Sounds interesting

[Joshua Clinard]

I would be glad if Jag added this to their plans.

[Vin DSL]

[CeleronXL]

Originally posted by Vin DSL

Ditto

[apgraham]

I'd even pay a bit extra for this. It doesn't have to be a really fast setup, just a presence.

-Al

[DIS]

Originally posted by Vin DSL

I agree

Now if we could find a solution to avoid email going down with the server that would be a good thig as well.

[Ron]

We need to think CAREFULLY about such a proposal.

First fo all, it's not simple to CORRECTLY implement this strategy, and may cause additional problems at every unavailability for a very long time.

Secondly, when you change your site contents, you'll have to do that virtually simultaneously with updating the failover database, or else it will assume that your site is down and implement it's strategy.

Third, when a web crawler visits your site and finds it to be unavailable, it (google, altavista, fast) assumes it to be temporary, and retries. If it were to find a "sorry" page, it would index that page, potentially discarding your entire site.

Fourth Zoneedit's particular system requires that you have TWO websites running to function. I realize the suggestion here is for a smaller scale/more simple system, but that simplicity is itself the cause of some of the complexity! lol

All of that said, one way to do this might be to utilize a service such as ZoneEdit, and point it to a page that will be redirected to a script which return either an http 307 or an http 503. I have not done research to see which would be preferable for what indexing agents. Nor am I certain that these would be the best choice, nor am I certain that this apprach is a good one. I should HOPE that whatever path you choose would be the best for GOOGLE

This is just off the top of my head. In a former life, I was in charge of failover/disaster recovery/business resumption for the unix systems for a large financial institution in Boston.

Just my 2 and one-half cent's worth, FWIW.

[lvzardoz]

Originally posted by Ron
Secondly, when you change your site contents, you'll have to do that virtually simultaneously with updating the failover database, or else it will assume that your site is down and implement it's strategy.

Third, when a web crawler visits your site and finds it to be unavailable, it (google, altavista, fast) assumes it to be temporary, and retries. If it were to find a "sorry" page, it would index that page, potentially discarding your entire site.

Fourth Zoneedit's particular system requires that you have TWO websites running to function. I realize the suggestion here is for a smaller scale/more simple system, but that simplicity is itself the cause of some of the complexity! lol

All of that said, one way to do this might be to utilize a service such as ZoneEdit, and point it to a page that will be redirected to a script which return either an http 307 or an http 503. I have not done research to see which would be preferable for what indexing agents. Nor am I certain that these would be the best choice, nor am I certain that this apprach is a good one. I should HOPE that whatever path you choose would be the best for GOOGLE

You make some very good points, Ron. One minor nitpick: my reading of Zoneedit information does not conclude that you need TWO websites (not including the error redirection one), I think that two is just the normal pattern when you are making a standard mirror-type failover.

As a possible alternative, in order to best satisfy Google, do you think a 307 or 503 error could be customized with URL parameters by using SSI or PHP to display a user's 'sorry' custom message?

[Ron]

It's been a while since I looked at this stuff, but I believe you can set the http__status_code (or whatever) using a php module... sorry I can't be more specific, but I'm not at my desk in my motorhome at the moment... I'm off visiting relatives.

Let me know.

-Ron.

[JPC-Greg]

Well a failover will be costly to implement. What if I start development on a new branch of jaguar technologies that caters to high end hosting. Fewer clients per machine, failover protection, email protection, and seperate dedicated techs for that operation ?

Of course that costs money to run so I have to poll to see who all would be interested. It goes without saying rates would be higher but for businesses and other people that want to protect their data Id say its worth it.

If a substantial number of clients want to move to such a division, at least enough to fund the operation, then I will put together a business plan for it and bring it to the surface this year.

[lvzardoz]

Originally posted by Jag
Well a failover will be costly to implement.

Actually, what I was talking about should not be costly at all. A simple one page non-graphics SORRY page accessed only when the main site is down should utilize little disk space and not much bandwidth.

In fact, for my own web sites I plan to implement my idea on my own using subdomains on BLINX.NET (100MB space, 1GB bandwidth, $12 per year)

I just need to do a bit more research on how it will impact Google rankings. My own inclination is to use a PHP "generated" 404 error with an embedded link so it still gets spidered (or links).

While it is true that a traditional failover solution requires full mirroring and that can be costly, I'm willing to settle for a partial solution at greatly reduced complexity and cost.

[G.Bloke]

Jag - I would be interested to see what such a high end plan and what the costs might be. Whilst the sites I host with Jaguar now aren't commercial I'm currently working on a commerical site.

Like all ventures it may or may not suceed and so paying for a dedicated solution would be premature and a lower end package suffers from all the stability issues. At least for piece of mind something between the two would be worth consideration.

[Vin DSL]

Excuse me, but I'm confused.

lvzardoz wants cheap failover. That sounded good.

The Chief comes back with how he can do it, but to do it right will cost you mo' money. I'm thinking, "Go girl! End of conversation." I figured he was being sarcastic.

Now it would seem this "simple/cheap failover strategy" is a "non-graphics SORRY page." WTF? What kinda strategy is that? I'd rather have ppl looking at a non-graphic IE error page than some generic "SORRY page."

Am I missing something here?

EDIT: My bad. Just re-read the original post. I'm changing my vote

[DIS]

Originally posted by Jag
Well a failover will be costly to implement. What if I start development on a new branch of jaguar technologies that caters to high end hosting. Fewer clients per machine, failover protection, email protection, and seperate dedicated techs for that operation ?

Of course that costs money to run so I have to poll to see who all would be interested. It goes without saying rates would be higher but for businesses and other people that want to protect their data Id say its worth it.

If a substantial number of clients want to move to such a division, at least enough to fund the operation, then I will put together a business plan for it and bring it to the surface this year.

This may be of interest Greg. Perhaps look at adding monitoring from an independant somewhere off Jags network into the mix.