6
votes
Help with Amazon Glacier (original) data retrieval
Hi folks. I'm in way over my head with an AWS problem. Mainly because I don't understand how to use AWS at all, but in 2013 I used a tool called FastGlacier to backup my data (including baby picture and videos) to Amazon Glacier. I also had them on a local drive and a HDD I gave to a friend. Alas, my drive failed a few years ago, and he lost the other in a move.
Anyway, I'm not a coder, and AWS baffles me. I'd like to retrieve these data but don't know where to start. Would anybody well versed in AWS be willing to walk me through the process? The AWS guidance is over my head.
Many thanks in advance.
It seems like the idea would be to restore the files you want to download, which makes them accessible through standard s3, and then you can download them with an s3 client like rclone. Take a look at the “AWS console” section on the linked help page.
I cannot fully walk you through the guidance, but I would be careful about how you access it. Pricing information
Direct transfer out of glacier is $.09, which could mean $90 for a terabyte of data.
It looks like transfer out to S3 in the same region is free, and bulk retrievals are free, so that's probably the route you want to go, then download it from S3 like @weldawadyathink said. You'll pay a much smaller amount of money for the S3 storage and transfer.
Honestly, if I knew nothing about it, I would ask Claude for instructions on how to do it, and emphasize you want it done 1) as cheaply as possible and 2) with no possible loss of data. I would tell it to fully document any commands it gave me with explanations. Then I would give the output to Gemini to validate. That will probably get you most of the way there, and if something doesn't work, you can give it the errors and keep asking it for guidance
Amazon is their own chat bot (Q I think) in the aws console, so you could try that too. It probably has more context for AWS services but is less advanced than Gemini or Claude.
I wonder how much LLMs are biased by "loss of data" since they aren't necessarily parsing exactly semantically. Something like "following data preservation standards" doesn't contain the token "loss" which may activate different numeric clusters. Probably doesn't make a difference 99% of the time but I do wonder about the 1% of cases.
Honestly, I'd be surprised if any of the frontier models had problems with a task of this scope, regardless of wording. But since it is sensitive, that's what the crosscheck is for
Glacier is cheap for day to day storage but expensive for retrieval so the bill will probably be more expensive than you expect. It may help to migrate to a specific storage class first (via a storage lifecycle mechanism?)
https://aws.amazon.com/s3/storage-classes/glacier/
Looks like S3 Glacier Flexible Bulk Retrieval or S3 Glacier Deep Archive are the less expensive options.
Also, Amazon will charge egress separately so it will help to use Direct Connect / or CloudFront if you can!