Proposal: Reorganize Consensus Node File System

Our current file system layout is extremely adhoc and the product of a long sequence of seemingly random decisions. I think we can do a much better job at organizing our file system.

Current Structure

Here, "root" is the current working directory of the JVM. (I'm 80% sure on this point... needs confirmation). I am not sure if there are other files elsewhere on the file system that the node utilizes.

root
├── VERSION
├── config.txt
├── data
│   ├── config
│   │   ├── api-permission.properties
│   │   ├── application.properties
│   │   ├── bootstrap.properties
│   │   ├── genesis.properties
│   │   └── node.properties
│   ├── diskFs
│   ├── keys
│   │   ├── private-node1.pfx
│   │   └── public.pfx
│   ├── onboard
│   ├── saved
│   │   ├── address_book
│   │   │   └── addressBookDebugFiles.txt
│   │   ├── com.hedera.services.ServicesMain
│   │   │   └── 0
│   │   │       └── 123
│   │   │           └── roundNumber
│   │   │               ├── SignedState.swh
│   │   │               └── variousStateFiles
│   │   ├── preconsensus-events
│   │   │   └── treeWithPcesFiles
│   │   ├── swirlds-recycle-bin
│   │   │   └── 0
│   │   │       └── recycledFiles
│   │   └── swirlds-tmp
│   │       └── temporaryFiles
│   ├── stats
│   │   └── legacyMetricsFiles.csv
│   └── upgrade
│       └── filesForUpgrade
├── eventStreams
│   └── events_0
│       └── eventStreamFiles
├── hedera.crt
├── hedera.key
├── log4j2.xml
├── output
│   ├── hgcaa.log
│   ├── queries.log
│   ├── swirlds-hashstream
│   │   └── hashStreamFiles.log
│   ├── swirlds-vmap.log
│   └── transaction-state
│       └── state-changes.log
├── settings.txt
├── settingsUsed.txt
└── stdout.log

Proposed Structure

Conventions:

• always prefer to use - in between words (as opposed to underscores or camel case)
• never include the node ID in any directory path beneath the root
• all files should be created with the file system manager, which will only permit files to be created within the data directory
• we should assume that for the time being that root will be a single mount point (i.e. it's safe to hard link between any two locations beneath root)
• the root directory should be configurable. All other directories below root should not be specified in configuration (old pattern is to make each directory configurable). We never change from the defaults, and having things configurable increases code complexity a lot. Better to use static constants for directory names beneath root.
• never use characters that are not compatible with all major operating systems
  • even if they are technically compatible, never use white space in file names
  • never use non-ASCII characters

root
├── bin
│   └── recycledFiles
├── data
│   ├── blocks
│   │   └── blockStreamFiles
│   ├── ces
│   │   └── consensusEventStreamFiles
│   ├── config
│   │   ├── VERSION
│   │   ├── api-permission.properties
│   │   ├── application.properties
│   │   ├── bootstrap.properties
│   │   ├── config.txt
│   │   ├── genesis.properties
│   │   ├── log4j2.xml
│   │   ├── node.properties
│   │   └── settings.txt
│   ├── keys
│   │   ├── hedera.crt
│   │   ├── hedera.key
│   │   ├── private-node0.pfx // currently there is an off by 1 error, can we remove the number entirely?
│   │   └── public.pfx
│   ├── logs
│   │   ├── address-book
│   │   │   └── addressBookDebugFiles.txt
│   │   ├── hgcaa.log
│   │   ├── queries.log
│   │   ├── settingsUsed.txt
│   │   ├── stdout.log
│   │   ├── swirlds-hashstream
│   │   │   └── hashStreamFiles.log
│   │   ├── swirlds-vmap.log
│   │   └── transaction-state
│   │       └── state-changes.log
│   ├── pces
│   │   └── treeWithPcesFiles
│   ├── snapshots
│   │   └── roundNumber
│   │       ├── SignedState.swh
│   │       └── variousStateFiles
│   ├── stats
│   │   └── legacyMetricsFiles.csv
│   └── upgrade
│       └── filesForUpgrade
└── tmp
    └── temporaryFiles