Skip to content

Latest commit

 

History

History
230 lines (159 loc) · 7.99 KB

README.md

File metadata and controls

230 lines (159 loc) · 7.99 KB
Raw

postcli

CLI tool for PoST initialization

Getting it

Go to the https://github.com/spacemeshos/post/releases and take the most recent release for your platform. In case if you want to build it from source, follow the instructions below.

git clone https://github.com/spacemeshos/post
cd post
make postcli

Alternatively you can use the postcli docker image:

docker pull spacemeshos/postcli:<version>

Replace the version with the latest version. You can find the latest version on the releases page. Avoid using the latest tag since it can be unstable/untested.

Usage

./postcli --help

Get OpenCL working

You need to have OpenCL support on your system. OpenCL usually comes with your graphics drivers. On Windows it should work out of the box on linux you will need to install them separately.

You can always list the providers by using

clinfo -l

That's separate command NOT shipped with post implementation. Please refer to your system installation manual of clinfo for installation instructions.

Nvidia

apt install nvidia-opencl-icd

AMD

https://amdgpu-install.readthedocs.io/en/latest/install-script.html

Intel

https://github.com/intel/compute-runtime/releases

Print the list of compute providers

./postcli -printProviders

Print the number of files that would be initialized

./postcli -numUnits 100 -printNumFiles
1600

Print the used config and options

./postcli -printConfig

Initializing PoST data

Example

./postcli -provider=2 -numUnits=4 -id=c230c51669d1fcd35860131e438e234726b2bd5f9adbbd91bd88a718e7e98ecb \
    -commitmentAtxId=c230c51669d1fcd35860131e438e234726b2bd5f9adbbd91bd88a718e7e98ecb -genproof

Remarks

  • Both -numUnits and -commitmentAtxId are needed to generate PoST data.

    • an already started initialization can be continued without providing these flags.

    • -commitmentAtxId: it is recommended to look up the highest ATX by querying it from a synced node with

      grpcurl -plaintext -d '' 0.0.0.0:9093 spacemesh.v1.ActivationService.Highest | jq -r '.atx.id.id' | base64 -d | xxd -p -c 64`

      The node can be operated in "non-smeshing" mode during synchronization and when querying the highest ATX.

      Alternatively you can pick an ATX from the current epoch from the explorer: https://explorer.spacemesh.io/epochs

  • If -id isn't provided a new identity will be auto-generated. Its private key will be stored in identity.key in -datadir next to the PoST data. This file then must to be copied/moved to the data/identities directory of the node managing this identity.

    • NOTE: The generated PoST data is ONLY valid for this identity!
    • If a public key is provided with the -id flag, the identities.key file will be NOT created. Make sure that the key file that belongs to the identity provided to postcli is available in the data/identities directory for a PoST service using the generated data to be able to connect to the node.

  • The -reset flag can be used to clean up a previous initialization. Careful: This will delete data that won't be recoverable.

Initializing a subset of PoST data

It is possible to initialize only subset of the files. This feature is intended to allow splitting initialization between many machines.

Example - split initialization between 2 machines

For this example we initialize 100 units and split the process of initialization into two chunks. This command shows the number of files that would be created during initialization:

./postcli -numUnits 100 -printNumFiles
1600

Note: Ensure that -id and -commitmentAtxId are the same for all subsets!

On machine A:

./postcli -numUnits 100 -id <id> -commitmentAtxId <id> -toFile 799 -datadir ./dataA

This will create the first 800 files in ./dataA directory. To verify:

ls -la ./dataA/*.bin | wc -l
800

On machine B:

./postcli -numUnits 100 -id <id> -commitmentAtxId <id> -fromFile 800 -datadir ./dataB

This will create the second 800 files in ./dataB directory. To verify:

ls -la ./dataB/*.bin | wc -l
800

Finally these sets can be combined by moving/copying the *.bin files into the same directory:

cp ./dataA/*.bin ./data/
cp ./dataB/*.bin ./data/
ls -la ./data/*.bin | wc -l
1600

Merging postdata_metadata.json: Every subset will create its own postdata_metadata.json. These files MUST be merged manually. During initialization a VRF nonce is searched that represents the index of the the label with the smallest value.

When postcli initializes in chunks, each subset can find a valid VRF nonce, which represents the local minimum in the initialized subset. It is necessary to manually select the best one (the global minimum) when merging the subsets.

The VRF nonces are stored as Nonce (index) and NonceValue (the label - 16 byte big endian numbers encoded in hex) in each post_metadata.json of every subset. Given two files:

... (other fields omitted for brevity)
"Nonce": 12345
"NonceValue": "0000ffda94993723a980bf557509773e"
... (other fields omitted for brevity)
"Nonce": 98765
"NonceValue": "0000488e171389cce69344d68b66f6b4"

The nonce in the second file (please see the NonceValue not Nonce field) is the global minimum since its value is smaller than the first one. The operator is required to find the smallest VRF nonce by hand and ensure that its index and value are in the postdata_metadata.json of the merged directory on the target machine.

Not every chunk will contain a VRF nonce in its postdata_metadata.json, but at least one should. If for the very unlikely case that no VRF nonce was found in any chunk the operator can run postcli again after merging the data without -fromFile and -toFile flags to find a VRF nonce.

Verifying initialized POS data

The postcli allows verifying an already initialized POS data. Verification samples a small fraction of labels from every file and compares them to labels generated with the same algorithm executed on CPU. Please note that generating labels on CPU is slow compared to GPU. Hence it is not possible to verify all the data (it would essentially mean re-initialization on CPU). If the GPU failed during initialization, the created PoST data will contain some or all invalid labels after that point. This method will only sample the PoST and might not detect a small amount of corrupted data.

Depending on PoST size and CPU speed a reasonable fraction (%) parameter needs to be picked that gives enough confidence but still completes verification in a reasonable time. Suggested values are <1%, closer to 0.1%.

To verify POS data:

  1. locate the directory of the POS data. It should contain postdata_metadata.json and postdata_N.bin files.
  2. run postcli -verify -datadir <path to POS directory> -fraction <% of data to verify>.

For example, postcli -verify -datadir ~/post/data -fraction 0.1 will verify 0.1% of data. No additional arguments (i.e -id) are required. The postcli will read all required information from postdata_metadata.json

If the POS data is found to be invalid, postcli will exit with status 1 and print the index of file and offset of the label found to be invalid. If verification completes successfully, postcli exits with 0.

Troubleshooting

Searching for a lost VRF nonce

In case you lost a VRF nonce after merging initialized subsets, you can use postcli to recover it without re-initializing the data. Postcli will need to read the entire POS data and find the nonce.

To find a lost nonce:

  1. locate the directory of the POS data. It should contain postdata_metadata.json and postdata_N.bin files.
  2. run postcli -searchForNonce -datadir <path to POS directory>.

The postcli will read the metadata from postdata_metadata.json and then look for the nonce in all postdata_N.bin files one by one. If the nonce is found it will update the metadata file.