Flexible Network Sandboxes

The binary tezos-sandbox uses the Flextesa library to provide sandbox networks with baker and endorser daemons, with various test scenarios.

Some of those scenarios run in the CI, see ./src/bin_sandbox/dune (although it is not recommended to experiment with sandboxes using dune which can leave unkilled processes easily, see issue #2445).


The application is not built by default, one needs:

make build-sandbox

(or make build-test).


See ./tezos-sandbox --help and all the examples below.

When running (semi-)interactive tests, it is recommended to wrap the call with rlwrap or ledit.

MacOSX Users

At runtime, sandboxes usually depend on a couple of linux utilities.

If you are on Mac OS X, you can do brew install coreutils util-linux. Then run the tests with:

export PATH="/usr/local/opt/coreutils/libexec/gnubin:/usr/local/opt/util-linux/bin:$PATH"

See Also

tezos-sandbox based on the Flextesa library which is being developed at gitlab.com/tezos/flextesa. One can find more instructions there, including how to use isolated sandboxes using Docker. TQ Tezos’ assets portal also shows how to start a sandbox and interact with it using a separate tezos-client: https://assets.tqtezos.com/docs/setup/2-sandbox/.


Let’s clear a couple of things up:

  • full Vs manual sandbox: we call “full” a sandbox that uses baker/endorser/accuser daemons and hence advances by itself. A “manual” sandbox only has nodes, they require successive calls to bake for tezos-client commands (or bake in the interactive prompt if any).

  • Each sandbox scenario has a root path where all logs and generated files go (usually exposed with the --root-path option).

  • Some sandboxes can start an interactive command-line interface. The command-line parsing uses the Sexplib library; see https://github.com/janestreet/sexplib#about for the lexical conventions. Try the help command for instance.

  • By default, sandboxed nodes get assigned successive port numbers for their RPC and P2P services; with the option --base-port PORT, PORT will be used for the RPC of node 0, PORT + 1 for its P2P, PORT + 2 for the RPC of node 1, etc.

  • Sandboxes like the mini-network also provide a shell-environment file at $ROOT_PATH/shell.env which provides aliases to tezos-client commands compatible with the sandbox (see also the help-env interactive command).


Interactive Mini-Network

One can run a mini-network advancing fast with accusers, bakers, and endorsers:

rlwrap ./tezos-sandbox mini-network \
       --root-path /tmp/zz-mininet-test \
       --size 2 \
       --number-of-bootstrap-accounts 2 \
       --tezos-node-binary ./tezos-node \
       --tezos-baker-alpha-binary ./tezos-baker-alpha \
       --tezos-accuser-alpha-binary ./tezos-accuser-alpha \
       --tezos-client-binary ./tezos-client

Once the network is started this test scenario becomes interactive:

Flextesa.mininet: Please enter command:

Just try h (or help) to see the available commands, or q to kill the sandbox and quit.

The mini-network has many options, ./tezos-sandbox mini --help.

Mini-Network with User Activated Upgrade

This example runs another full sandbox (3 nodes, 2 “bakers”), for a limited amount of time (60 blocks, no interactivity), and performs a user-activated-upgrade (a.k.a. a protocol “hard-fork”) between Babylon and Carthage (as built on the master branch).

We also set the base-port to 3000 and add some random traffic; i.e. create contract originations and contract calls (for now the setting also requires also the --until-level option):

./tezos-sandbox mini-network \
       --root-path /tmp/hard-fork-mininet \
       --size 3 \
       --base-port 3_000 \
       --number-of-bootstrap-accounts 2 \
       --protocol-hash PsBabyM1eUXZseaJdmXFApDSBqj8YBfwELoxZHHW77EMcAbbwAS \
       --protocol-kind Babylon \
       --until-level 60 \
       --random-traffic any \
       --tezos-baker-alpha-binary ./tezos-baker-005-PsBabyM1 \
       --tezos-endorser-alpha-binary ./tezos-endorser-005-PsBabyM1 \
       --tezos-accuser-alpha-binary ./tezos-accuser-005-PsBabyM1 \
       --hard-fork 20:PsCARTHAGazKbHtnKfLzQg3kms52kSRpgnDY982a9oYsSXRLQEb \
       --hard-fork-baker-alpha-binary ./tezos-baker-006-PsCARTHA \
       --hard-fork-endorser-alpha-binary ./tezos-endorser-006-PsCARTHA \
       --hard-fork-accuser-alpha-binary ./tezos-accuser-006-PsCARTHA \
       --tezos-node-binary ./tezos-node \
       --tezos-client-binary ./tezos-client

Manual Mini-Network With An Archive Node

An interactive Carthage sandbox with 3 nodes, one of which running in archive mode, and no baking daemons:

rlwrap ./tezos-sandbox mini-network \
       --root-path /tmp/manual-mininet \
       --size 3 \
       --set-history-mode N000:archive \
       --no-baking \
       --protocol-hash PsCARTHAGazKbHtnKfLzQg3kms52kSRpgnDY982a9oYsSXRLQEb \
       --protocol-kind Carthage \
       --tezos-baker-alpha-binary ./tezos-baker-006-PsCARTHA \
       --tezos-endorser-alpha-binary ./tezos-endorser-006-PsCARTHA \
       --tezos-accuser-alpha-binary ./tezos-accuser-006-PsCARTHA \
       --tezos-node-binary ./tezos-node \
       --tezos-client-binary ./tezos-client

Once the network is started, we enter the interactive mode, and we can use the bake command to create blocks (bake can take an argument: which client to bake with, e.g. bake 2).

We can check that N000 is indeed an archive node: c2 rpc get /chains/main/checkpoint (help tells us that c2 is the client for the node N000).

Double Endorsement Accusation

There are 3 “accusation scenarios” so far, see ./tezos-sandbox accusation --help. For instance, the following command starts a small 3-node network, forces one baker to endorse two concurrent branches, and then makes another baker inject (and bake) the double-endorsement-evidence operation. The option --pause-at-end=true tells tezos-sandbox to enter the interactive mode (command prompt) at the end of the test to give a chance to explore the sandbox before killing all the nodes.

PATH=.:$PATH rlwrap ./tezos-sandbox accusations simple-double-endorsing \
     --root $PWD/double-endorsing-test \

Voting With a Ledger Nano Device

The voting test tries to do a full round of voting and protocol switch, including baking on the test-chain, see documentation in ./tezos-sandbox voting --help.

The test can run in a simpler-to-setup, or “degraded,” mode of operation (cf. call in ./src/bin_flextesa/dune for the version which run in Gitlab-CI pipelines). In this example, we run instead a full test with a Ledger Nano device as one of the bakers/voters. The test automatically becomes interactive because the user has to press buttons on the device, including for changing between apps.

Get an URI for your ledger (the test requires both the Wallet and Baking apps):

tezos-client list connected ledgers

And use the URI (no need to import it) for the --with-ledger option:

rlwrap ./tezos-sandbox voting \
     ./src/proto_006_PsCARTHA/lib_protocol/TEZOS_PROTOCOL \
     ./src/proto_demo_noops/lib_protocol/TEZOS_PROTOCOL \
     --with-ledger "ledger://crouching-tiger-hidden-dragon/ed25519/0'/0'" \
     --serialize-proposals \
     --base-port=20_000 \
     --current-node-binary ./tezos-node \
     --current-client-binary ./tezos-client \
     --winner-client-binary ./tezos-client \
     --current-admin-client-binary ./tezos-admin-client \
  • The first path argument has to be the path to a valid protocol which can be switched to from the current (proto_alpha) one.

  • The second protocol, the looser, only needs to be valid for the protocol compilation.

  • The option --serialize-proposals tells the test to call tezos-client submit proposals for ... one proposal at a time which is the only method the ledger Wallet app can really understand.

  • The *-binary options allow to set the paths to the executables for the different protocols: current and winner.

The test becomes interactive and guides you through the interactions with the ledger, e.g.:


      Setting up "ledger://crouching-tiger-hidden-dragon/ed25519/0'/0'" for
      baking. The ledger should be showing the setup parameters (Address,
      Main chain, HWMs).

     Please hit “✔” on the ledger.

Implementation Considerations

Running_processes is very high-level (actually agnostic to Tezos). Most processes are actually calls to sh -c <script> where <script> is the result of a Genspio compilation, this leaves the option to later easily run some processes over SSH (without OCaml dependencies on the destination host) or in special containers (e.g. docker run --cpu-shares ...).

The prompt commands for interactive use use Base.Sexp.t parsers (because already a dependency, and we need a good string literal parser so we cannot use Clic nor Cmdliner).

Special Coding Style

A fresh “just for testing project” is a good occasion to experiment a bit …

See ./vendor/lib_flextesa/internal_pervasives.ml:

  • EF: we try to use combinators on top of Easy-format for most pretty-printing (it is still compatible with Format but it is much more functional/composable and does not rely on @[<2,3>@{crazy}@ @<acronym>EDSLs@n@]).

  • Many standard modules are taken from Jane St Base (already a dependency of Tezos): List, String, Option, Int, Float.

  • Error monad uses more typed errors (polymorphic variants), cf. module Asynchronous_result (and note that bind also calls Lwt_unix.auto_yield 0.005 ()).

  • All state is kept in a (non-global) value passed as argument everywhere needed. To simplify the dependency management the state variables are objects (cf. Base_state, then Paths, Console, etc).