Skip to content

A datastore implementation using sharded directories and flat files to store data

License

Notifications You must be signed in to change notification settings

ipfs/go-ds-flatfs

Repository files navigation

go-ds-flatfs

standard-readme compliant GoDoc Build Status Coverage Status

A datastore implementation using sharded directories and flat files to store data

go-ds-flatfs is used by go-ipfs to store raw block contents on disk. It supports several sharding functions (prefix, suffix, next-to-last/*).

It is not a general-purpose datastore and has several important restrictions. See the restrictions section for details.

Lead Maintainer

Jakub Sztandera

Table of Contents

Install

go-ds-flatfs can be used like any Go module:

import "github.com/ipfs/go-ds-flatfs"

Usage

Check the GoDoc module documentation for an overview of this module's functionality.

Restrictions

FlatFS keys are severely restricted. Only keys that match /[0-9A-Z+-_=]\+ are allowed. That is, keys may only contain upper-case alpha-numeric characters, '-', '+', '_', and '='. This is because values are written directly to the filesystem without encoding.

Importantly, this means namespaced keys (e.g., /FOO/BAR), are not allowed. Attempts to write to such keys will result in an error.

DiskUsage and Accuracy

This datastore implements the PersistentDatastore interface. It offers a DiskUsage() method which strives to find a balance between accuracy and performance. This implies:

  • The total disk usage of a datastore is calculated when opening the datastore
  • The current disk usage is cached frequently in a file in the datastore root (diskUsage.cache by default). This file is also written when the datastore is closed.
  • If this file is not present when the datastore is opened:
    • The disk usage will be calculated by walking the datastore's directory tree and estimating the size of each folder.
    • This may be a very slow operation for huge datastores or datastores with slow disks
    • The operation is time-limited (5 minutes by default).
    • Upon timeout, the remaining folders will be assumed to have the average of the previously processed ones.
  • After opening, the disk usage is updated in every write/delete operation.

This means that for certain datastores (huge ones, those with very slow disks or special content), the values reported by DiskUsage() might be reduced accuracy and the first startup (without a diskUsage.cache file present), might be slow.

If you need increased accuracy or a fast start from the first time, you can manually create or update the diskUsage.cache file.

The file diskUsage.cache is a JSON file with two fields diskUsage and accuracy. For example the JSON file for a small repo might be:

{"diskUsage":6357,"accuracy":"initial-exact"}

diskUsage is the calculated disk usage and accuracy is a note on the accuracy of the initial calculation. If the initial calculation was accurate the file will contain the value initial-exact. If some of the directories have too many entries and the disk usage for that directory was estimated based on the first 2000 entries, the file will contain initial-approximate. If the calculation took too long and timed out as indicated above, the file will contain initial-timed-out.

If the initial calculation timed out the JSON file might be:

{"diskUsage":7589482442898,"accuracy":"initial-timed-out"}

To fix this with a more accurate value you could do (in the datastore root):

$ du -sb .
7536515831332    .
$ echo -n '{"diskUsage":7536515831332,"accuracy":"initial-exact"}' > diskUsage.cache

Contribute

PRs accepted.

Small note: If editing the README, please conform to the standard-readme specification.

License

MIT © Protocol Labs, Inc.

About

A datastore implementation using sharded directories and flat files to store data

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages