Skip to main content

Object storage (S3)

Object storage is separate from your environment's NVMe disks. It is reached through our own S3 endpoint — not the storage provider's. Nothing about the provider appears in your configuration, so we can move a bucket to different hardware later without changing your endpoint, bucket name or keys.

You choose one of two ways to pay for it, and the choice belongs to the bucket:

  • a reserved allocation of capacity in 10 GB blocks, charged whether you fill it or not; or
  • pay for what you store, which buys no capacity up front and charges for how much the bucket holds.

Every bucket is private. Another client cannot read, write or even list it, and their credentials are refused at our endpoint.

The price comes from the price book

Before you confirm, the panel shows what the capacity costs per month including VAT, what the rest of this month costs after proration when you are enlarging, the downloads included each month and the rate for additional downloads. Every figure comes from the same price book your invoice is produced from, so the two cannot disagree. If no price is recorded for the product, the panel says so instead of showing a number.

Choosing hosting capacity​

For applications and databases, start with choosing your hosting size. We recommend Professional (2×) for most projects; Premium (3×) adds extra headroom, while Essentials (1×) suits lighter workloads. Object storage below is a separate purchase: it adds space for objects, not CPU or memory for your applications.

Standard monthly prices from 1 November 2026​

The live price list is at hosting.haak.co/pricing — that page is read from our price book each time it loads, so it is the current figure. The table below records the prices as settled for 1 November 2026 and is kept for the record.

Prices below include VAT. Your reviewed quote is authoritative: a company agreement may set a different price, and a partial month is prorated rather than charged as a full month.

ItemFull-month price including VAT
One standard hosting unit (Essentials · 1×), including 40 GiB NVMeR650.00
Additional NVMe, per 10 GB block, two stored copies (arranged with us)R25.00
S3 capacity, per 10 GB, reserved allocationR3.00
S3 capacity, 100 GB, reserved allocationR30.00
S3 charged for what you store, per GB per monthR0.30
One standard unit plus 100 GB S3R680.00
Additional measured downloads, per 10 GB blockR1.00

The allocations stay separate: 40 GiB NVMe plus 100 GB S3, not one combined disk. A reserved allocation includes monthly downloads matching its purchased capacity; a bucket that pays for what it stores includes downloads matching what it actually stored over the month. Platform reads are recorded separately and excluded from your download charge. You can set a monthly download-overage spending ceiling in Storage → Downloads. Your overage charge cannot exceed it; if traffic goes above that limit, Hosting absorbs the difference. Admission may refuse further external downloads at the limit, but the billing ceiling applies even if admission is unavailable. Unproved usage is shown as unknown, not silently counted as zero.

Two ways to pay​

Reserved allocation works like buying a disk: you choose a capacity in whole 10 GB blocks, and that whole capacity is charged for each month you hold it, whether you fill it or not. A managed directory mount needs a reserved allocation, because that is what the mount is billed against.

Pay for what you store is for buckets you reach with an S3 client or SDK. The bucket is created without a capacity purchase and charged for what it holds, so it grows as your application writes and shrinks — in cost, not in data — as you delete. There is nothing to enlarge and no purchase to confirm.

You can move a bucket between the two later and it keeps every object, key and endpoint. A bucket must be detached from every managed directory before it can pay for what it stores; moving the other way needs an allocation large enough to hold what the bucket already stores. Do it with hostingctl storage mode <bucket> --native (or --allocated), or ask us.

Buy storage​

  1. Open your company and choose Storage → Object storage, then Add object storage.
  2. Enter a name you will recognise. For a reserved allocation, enter the capacity in GB, which must be a whole number of 10 GB blocks. A bucket that pays for what it stores needs no capacity — ask us to set one up.
  3. Confirm. The bucket is created for you — there is no provider console, no ticket and no setup step.

While it is being set up the bucket says Setting up. When it says Ready, its connection details on Storage → Object storage are usable. If it says Failed, the platform is retrying. Contact support if it stays there.

How a bill is worked out​

Reserved allocation is the simple one: the price of each 10 GB block times the blocks you hold, prorated from the moment the storage became usable. A partial month is charged for the part you held, not as a full month.

Pay for what you store is measured, not sampled. Every hour the platform lists the bucket and adds up what it holds. Each hour's figure is rounded up to the next whole GB and weighted by how long it applied, and the month's charge is those GB-hours at R0.30 per GB per month. Two consequences are worth knowing:

  • a bucket holding 1,200 GB for a full 31-day month costs R360.00 including VAT, the same as reserving 1,200 GB;
  • 100.1 GB counts as 101 GB every hour, so a constant month of it costs R30.30, not R30.03.

There is no monthly minimum, no peak-hour price and no charge for an empty bucket. Deleting objects lowers the hours that follow; it cannot reduce the hours already recorded, because they happened.

What "unmeasured" means

If an hour could not be measured — the listing failed, or it was too large to finish in the time allowed — the platform reports that stretch as unmeasured and holds the storage charge for it. Unmeasured is not the same as empty and not the same as zero: a month we could not prove is not charged, and it is shown as unknown rather than as a nil bill. We re-measure the hour, and the charge appears once the meter can prove it. Download measurement follows the same rule.

Connect an S3 client​

The bucket's page shows the four values every S3 client needs:

SettingWhat to use
EndpointThe address shown, for example https://s3.example
BucketThe bucket name shown, for example hsk-…
RegionThe signing region shown. It is a fixed label; the storage is not in that cloud region
AddressingPath style — the bucket name is part of the URL path

And one key pair:

  • Issue your first key with Rotate access key. A bucket is provisioned without handing you a secret, so the rotation answer is where the pair is shown; the bucket page keeps only its last four characters afterwards.
  • the access key ID is shown in full when it is issued and afterwards by its last four characters;
  • the secret access key is shown once, in the answer that issues it. Copy it then. If you lose it, rotate the key and use the new one — the old secret cannot be retrieved.

With the AWS CLI, for example:

export AWS_ACCESS_KEY_ID="<the access key ID>"
export AWS_SECRET_ACCESS_KEY="<the secret, copied when it was issued>"
export AWS_DEFAULT_REGION="<the signing region>"

aws --endpoint-url https://s3.example s3api list-objects-v2 --bucket hsk-yourbucket
aws --endpoint-url https://s3.example s3 cp ./video.mp4 s3://hsk-yourbucket/assets/video.mp4
aws --endpoint-url https://s3.example s3 cp s3://hsk-yourbucket/assets/video.mp4 ./copy.mp4

Most clients need path-style addressing switched on explicitly; a client that insists on virtual-hosted style (bucket.endpoint) will not work with this endpoint.

Manage objects​

The panel lists your keys a page at a time and can:

  • list a prefix, so you can check what an application wrote;
  • upload a file, and download one, using a short-lived signed link;
  • delete one named key, after you confirm it.

Object bytes travel directly between your browser or client and our endpoint. They do not pass through the control panel, and uploaded content is never executed here.

For anything large or scripted, use a normal S3 client — that is the supported path, and the panel's listing is a convenience rather than a bulk browser.

Change capacity​

Reserved allocation. Choose Add capacity and enter the new total. Capacity grows in whole 10 GB blocks through the priced quote you confirm. A reserved allocation can also be reduced in whole 10 GB steps, but only against a recent measurement: if the last reading is old, or the bucket already stores more than the target, the reduction is refused rather than accepted blindly. Use hostingctl storage resize <bucket> --size <gb>, or ask us. It changes what you are billed and never deletes an object, so your data stays until a deletion is explicitly ordered.

Pay for what you store. There is no capacity to change. The bucket grows as your application writes and the bill follows; nothing needs enlarging first.

Beside what a bucket is set to, the panel shows what was last measured. A measurement is a reading, not a live figure:

  • unknown means nothing has been measured yet — that is not the same as empty;
  • stale means the last reading is old, so the bucket may hold more or less now.
Capacity is not a hard write quota

Neither charging basis makes the endpoint refuse writes when a bucket passes a size. On a reserved allocation, exceeding what you bought will not stop an application — and it does not silently become free either. On pay for what you store there is no size to exceed, and the bill simply follows what you keep. Watch the measured figure, and for a reserved allocation treat the allocation as a budget rather than a ceiling.

Access keys​

  • Rotate issues a new key pair and keeps the previous key working, so you can update your configuration without downtime. The new secret is shown once. Revoke the old key when the new one is in use — a bucket holds at most two keys at a time.
  • Revoke removes a key immediately; anything still configured with it stops working. While an environment mounts the bucket, revoking is refused: a running mount keeps the key it started with, so detach that environment first (detaching releases the mount and a rebind picks up the remaining key), then revoke. Native access has no such limit — update the application's configuration, then revoke.
  • Every key is scoped to its own bucket. A key cannot be used against another bucket, and it cannot list the buckets in the account.

Use it from an environment​

A bucket can be bound to one of your environments so the platform records which environment reaches it:

  • Native S3 — the application connects with the endpoint and a key pair above. This is available now and is the normal choice for an application with an S3 plugin or SDK.
  • Managed directory mount — the platform mounts the bucket at an absolute directory inside the application, for example /var/www/html/wp-content/uploads, so the application writes ordinary files and they land in the bucket. Choose it when the application has no S3 support or when a plugin already writes to a directory. A directory can belong to one bucket per environment, and it replaces the bundled storage for that path — copy existing files across before relying on it. Uploads from the mount happen asynchronously: a file you close reaches your bucket shortly after, but closing it does not wait, so an application that must be certain the bytes are stored has to call fsync and check that it succeeded. If the node's mount process is lost before that upload completes, a write the application considered finished can be lost.

Binding records intent; it does not copy data and does not change what you pay. Detaching an environment removes that access path and touches no object. A retired bucket stops serving object access from this panel too, because every operation signs with your own key rather than ours.

Rotate a storage encryption key​

For a bucket with encryption enabled, open Storage → Object storage → Encryption key and choose Rotate encryption key. You need permission to manage the company's storage and strong authentication. Review the confirmation before starting. This changes the key protecting stored data; your applications keep their existing endpoint, bucket name and access keys.

The request starts background work. New writes use the replacement key while existing objects are re-encrypted, and storage remains available. The bucket shows the phase and number of objects processed. Only one rotation can run at a time. If retained object versions still need the previous key, Hosting keeps it until those versions expire under their retention policy. A retry resumes existing work.

You can also start a rotation and read its progress with hostingctl. Neither surface returns the encryption key.

Protection, deletion and cost​

  • Object content is private to your company and travels over TLS. Bucket encryption at rest is not a self-service option today: we can seal an eligible bucket at rest, but the option is not switched on for self-service purchases. If your objects need encryption at rest, tell us before you buy — it is arranged per bucket, and objects written before that arrangement are not converted.
  • The storage provider keeps redundant copies, which protects against a disk failure. That is not a backup: there is no versioning history and no point-in-time restore of objects. Keep your own copy of anything you cannot lose.
  • Deleting an object removes it permanently. Deleting a bucket's access (retiring it) keeps the objects; removing the data is a separate request.
  • The download terms include a monthly allowance, then the quoted overage rate. A reserved allocation earns allowance from its reserved capacity, prorated for the time it held that capacity. A bucket that pays for what it stores earns allowance from its stored data, averaged over the month. The panel shows which buckets were measured for the month. Only measured buckets can incur download charges. You may set a spending ceiling for extra downloads: it is the most you can be charged for download overage in a month, and any amount above it is ours to carry. While download charging is not yet enabled, no download charge appears on an invoice.

In the Downloads tab, find your bucket and read Allowance remaining for the selected month. Pending measurement identifies figures still awaiting a full download measurement. Traffic recorded so far is the partial amount available. A measured bucket can show 0 GB, including when it has used all of its allowance.

Backup bucket​

Company → Storage → Backup destination holds Your own backup bucket. Backups already go to our bucket, which our retention promises cover. Pointing them at your own S3-compatible bucket as well gives you a second copy that you hold and can expire yourself. It is never the copy we restore from.

Fill in the Endpoint, Bucket, Region and, optionally, the Folder inside the bucket, together with the Access key ID and Secret access key. Both credentials are asked for on every save, because neither is ever shown back to you — so we cannot fill them in for you.

Start copying to it now decides whether anything is written to the bucket yet. Left unticked, the destination is recorded and copies nothing until you tick it. Choose Save destination to record it, and Remove destination to stop writing new copies; the copies already written stay in your bucket, and our own copy and its retention are unaffected.

What actually arrives in the bucket, how often, and how to open it — including the restic repository password for the file archive — is on Backups and restores.