Volumes
Disk that outlives a copy. Where it comes from, what it survives, browsing it, and backing it up.
Anything a running copy writes to its own filesystem is lost when that copy restarts. A volume is a directory that is kept.
On ISOGrid you do not create volumes by hand. They come with what you deploy. This page says where they come from, what each one survives, and what you can do with them.
Volume or store
A volume is a directory. Use it for software that expects files on disk and cannot be changed: a CMS with an uploads folder, a tool that reads its plugins from a path.
A store is reached over the network with S3, by any number of applications, from anywhere. Use it when you write the code yourself. It is the better home for files that many copies serve, and for files that must be reachable from outside the application. See Object storage.
Rows belong in a database, not in either.
Where a volume comes from
An application's disk comes with its size. Some instance types include disk,
shown in GB beside the CPU and memory. Choose one and the application gets a
single directory, mounted at /data. That is the only way to get one. You
cannot add a second, or choose the path, and the volumes in a stack or compose
file are not created. The import tells you so rather than dropping them
silently.
A solution's files come with the solution. WordPress, Odoo, n8n and the other ready-made solutions keep your own files on a volume created with them: themes, plugins, uploads, add-ons, workflows.
Databases and managed services keep their data on disk too, but that disk is the engine's, not yours to open. It is not listed.
What it survives
An application's disk is kept when the application restarts, is redeployed, is scaled, or moves to another machine. Every copy of the application sees the same directory.
Three limits come with that:
- Nothing arbitrates between copies. Two copies writing the same file at the same moment will damage it. An embedded database file on a shared disk needs exactly one copy.
- The disk is kept in one place in the region and is not replicated. More copies protect the application, not its files. While the storage is unavailable, the application cannot read them.
- The size is a hard limit. When the disk is full, writes fail.
Moving to an instance type with more disk grows it on the next deployment and keeps what is there. It does not shrink: do not move an application that holds files to an instance type with less disk.
Application storage is set up per region. Where a region does not have it yet,
the application still deploys, without the disk, and the deployment log says so.
Read that line before you trust /data with anything.
A solution's files live on the one machine the solution was created on, and the solution stays on that machine. They are kept across a restart, a stop and start, a resize and a redeploy. If that machine is down, the solution is down until it returns.
Looking inside
The Volumes page lists the volumes your running applications use, with the application and the path each is mounted at. An application's own page shows the same for that application, and a solution has a files view.
Browsing is read-only. You can list directories and open files. You cannot upload, edit, rename or delete, in the console or through the API. Change files through the application itself.
What to expect:
- Only running things are listed. A stopped application has no copy to read from.
- A file is shown up to its first 512 KB, and a directory up to its first 1,000 entries. The page says when it stopped short.
- A file that is not text is shown as a rough preview, not as its contents.
- For a solution, only your own files are shown. The engine's own data is kept elsewhere and cannot be reached by naming it.
Backups
A solution's files can be backed up from its Backups view. A backup is a copy of the whole directory, taken elsewhere. It is off by default.
- On demand. Back up now takes one copy and does not start a schedule.
- On a schedule. Every hour, every day, every week or every month. For anything but hourly you choose the hour, in UTC. A monthly backup can be set up to the 28th, so that it also runs in February.
- Kept for at most three days. That is a ceiling, not a default you can raise. A copy of a whole directory is a safety net for a bad plugin or a mistaken deletion, not an archive. Download the copy you want to keep longer.
- Turning the schedule off deletes nothing. Copies already taken stay until they age out.
A backup copies that directory and nothing else. A database the solution uses is not part of it.
A copy taken while the solution is running may come with a note that files changed while it was being read. It is still usable, but it is a slightly blurred moment rather than an exact one.
Restoring is destructive. The directory is emptied and the copy is put in its place. Everything written since that copy was taken is lost. Stop the solution first: a restore is refused while it is running, because restoring underneath a running site leaves it half on the copy and half on what was there.
For an application's disk, the console offers browsing only. If its files matter, keep a second copy somewhere your application writes to, such as a store.
Deleting
There is no separate delete for a volume. It goes with what owns it.
Deleting an application deletes its disk and every file on it. A redeploy, a restart or a change of scale never does.
Deleting a solution deletes its files and its database. There is no undo.
Download any backup you want to keep before you delete. Do not count on copies outliving the thing they were taken from.
Edge cases worth knowing before you meet them
The Volumes page is empty. Nothing with a volume is running. Start the application and look again.
An image declares a volume of its own. It appears in the list while the application runs, but it belongs to one copy and is not kept: a new copy starts with an empty one. Only the disk that comes with the instance type is kept.
A row is marked as a bind mount. It is a path on the machine rather than a volume. It is listed and cannot be opened.
The application wrote a file and you cannot find it. It was written outside
/data, into the copy's own filesystem, and went with that copy.
What to read next
- Object storage for files shared over S3.
- Deploying an application
- When things go wrong