Post AYhrk2dDDObHQ8Feee by jonathanmatthews@fosstodon.org
(DIR) More posts by jonathanmatthews@fosstodon.org
(DIR) Post #AYhnb6ikF5hNgEK8dk by simon@fedi.simonwillison.net
2023-08-14T01:00:41Z
0 likes, 0 repeats
Anyone got thoughts on good backup strategies for data where the only copy is in an S3 bucket?I worry about things like catastrophic billing failures or some weird ML algorithm deciding to ban an AWS account for some reasonBackup to a separate account in another region? Mirror bucket contents to a server on another hosting provider?
(DIR) Post #AYhnt7oMvStoH8Xtz6 by Melaskia@mastodon.social
2023-08-14T01:03:53Z
0 likes, 0 repeats
@simon I guess I would look at the cost of transferring the data from one bucket to another, or to another provider. You would need to pay potentially for egress and ingress depending on what kind of solution you go for.
(DIR) Post #AYho69RcAKnV6Y78hE by jesse@metasocial.com
2023-08-14T01:04:04Z
0 likes, 0 repeats
@simon I’d probably want to mirror off amazon, to prevent “amazon decides to stop doing business with you” attacks. Last I looked, Digital Ocean’s pricing was decent.
(DIR) Post #AYhoGuziENONTSCHwm by webology@mastodon.social
2023-08-14T01:04:19Z
0 likes, 0 repeats
@simon between Backblaze, Cloudflare, and Digital Ocean, all three have s3 compatible services (most are even cheaper) than s3 that might be cheap backups for you.
(DIR) Post #AYhoGwmfZq9D1a6ttY by webology@mastodon.social
2023-08-14T01:05:56Z
0 likes, 0 repeats
@simon it sounds like you might want separate AWS accounts if you don't already have them. That might give you flexibility that if the main service account gets banned, you don't lose other unrelated services.
(DIR) Post #AYhoR6eHkrmAW126IC by cloudthethings@hachyderm.io
2023-08-14T01:05:27Z
0 likes, 0 repeats
@simon Backblaze and Cloudflare both support S3-compatible storage at very competitive prices so you could use either (or both) of those with any S3 replication/cloning tool. You could even fail over to serve direct from Cloudflare in event of an AWS issue: not sure of read costs on Backblaze.
(DIR) Post #AYhof9XJocjhWdTKrI by jonoabroad@mastodon.nz
2023-08-14T01:06:52Z
0 likes, 0 repeats
@simon backblaze b2 copy?
(DIR) Post #AYhp55dyuVVkywxMUS by krusynth@mastodon.publicinterest.town
2023-08-14T01:08:45Z
0 likes, 0 repeats
@simon every place I’ve worked at we just used the built-in replication honestly - including my current role.
(DIR) Post #AYhpGDhNIsDB9HqoZk by while1malloc0@hachyderm.io
2023-08-14T01:09:29Z
0 likes, 0 repeats
@simon A company I worked for backed up to a bucket in a disaster recovery account, but since IIRC all AWS accounts are associated with a billing account, that might not help against a catastrophic billing failure unless there’s also a backup billing account.
(DIR) Post #AYhpUdCRHufIIEFjTU by scanner@apricot.social
2023-08-14T01:15:38Z
0 likes, 0 repeats
@simon I mirror with other providers and... I keep a local copy at home. (I mean.. terabytes of local disk is pretty easy.. but you want it backed up.. ) (I use Backblaze B2 and Digital Ocean, and minio on machines in my home.)
(DIR) Post #AYhq9iertjmx5y8tto by developerjustin@mastodon.social
2023-08-14T01:29:12Z
0 likes, 0 repeats
@simon you could replicate to an S3-compatible service like B2 from Backblaze.
(DIR) Post #AYhqrxGawfDN90skO8 by portertech@social.portertech.ca
2023-08-14T01:37:08Z
0 likes, 0 repeats
@simon I mirror to a self-hosted minio, a fantastic piece of software
(DIR) Post #AYhrk2dDDObHQ8Feee by jonathanmatthews@fosstodon.org
2023-08-14T01:47:01Z
0 likes, 0 repeats
@simon GCP’s Storage Transfer Service to a GCS bucket with appropriate storage class settings. It can run periodically, and it’s totally hands off and serverless. It’d cost you the S3 transfer + API call fees plus GCS storage fees.
(DIR) Post #AYhsmrg0V9r6GdT8UK by leadegroot@bne.social
2023-08-14T01:58:39Z
0 likes, 0 repeats
@simon backup to glacier storage (in a different region)?Cheapest way to maintain something you hope to never access?
(DIR) Post #AYhvHOM5lbmuDMq17g by thejodie@mstdn.social
2023-08-14T02:26:44Z
0 likes, 0 repeats
@simon use rclone to sync to other providers/hosting/on-prem.
(DIR) Post #AYhwXmBV2R2ehtGBLk by cboettig@ecoevo.social
2023-08-14T02:40:44Z
0 likes, 0 repeats
@simon second the suggestion of self-hosted MINIO (open source but same S3 API). Also true of redhat CEPH, used by lots of academic data centers including National Science Foundation’s openstoragenetwork.org
(DIR) Post #AYi2ie1gJJtOJZEWhs by AbeEstrada@mastodon.social
2023-08-14T03:49:58Z
0 likes, 0 repeats
@simon https://www.arqbackup.com
(DIR) Post #AYi9BZ7UEN89Dk7eGu by jszym@cosocial.ca
2023-08-14T05:02:23Z
0 likes, 0 repeats
@simon I would likely rclone to another provider like B2 or R2.If the size of the bucket is manageable, I might rclone to a server where I could run Borg or restic to get real backups.You can never get guarantees from the powers that rule the server firmament, but rclone is made for this stuff so it likely fits your requirements.
(DIR) Post #AYi9N5tihfxW46BFa4 by steve@deliverabilit.ie
2023-08-14T05:02:36Z
0 likes, 0 repeats
@simon A second independent copy on backblaze B2 would defend against what you’re concerned about. And it’s just another S3 compatible system so it should be easy enough to do.Glacier would defend against other sorts of failure. Maybe both?
(DIR) Post #AYi9eQqi3lITy42WdU by pandora@chaos.social
2023-08-14T05:07:45Z
0 likes, 0 repeats
@simon if u use a solution like duplicati u can select multiple backup places including local and cloud storages
(DIR) Post #AYiVC4vw97jOsfMzh2 by medecau@hachyderm.io
2023-08-14T09:08:55Z
0 likes, 0 repeats
@simon catastrophic billing failure probably requires 1) alt payment method to be in place before catastrophe hits, 2) some solution on the financial side, think alt payment from different bankweird ML algorithm probably requires 1) strong KYC state with service provider, 2) alt service provider, one with whom you also have good KYC standing with
(DIR) Post #AYio4XC15pEHL10Vii by brunomvsouza@hachyderm.io
2023-08-14T12:40:28Z
0 likes, 0 repeats
@simon my important backups (e.g. Google Photos' accounts from my family) are stored in two different providers, each with accounts created with different domains I “own” in different registars.To keep the costs low, I carefully pick storage tiers based on the durability and the likelihood of needing to restore data from that provider, I usually anticipate reading data from these backups primarily in disaster recovery scenarios.I mirror them incrementally with rclone.
(DIR) Post #AYkEL2MTlfkYbcOHTs by laurencerowe@mastodon.cloud
2023-08-15T05:09:28Z
0 likes, 0 repeats
@simon I ended up mirroring to GCP when this was a worry.