Repository navigation
test: sync operators on a four-node BFT synchronizer - #63
Conversation
Adds a -B LocalNet profile that runs the app-synchronizer on four BFT nodes, each run by its own org with its own participant and sync operator. At startup the four orgs create the operator party together: they own its namespace and host it on their participants. A fifth node and its org come up outside the synchronizer. LocalNetDedicatedSyncBftIntegrationTest checks that purchases and consumption agree on every node and org, that node 5 and its org join and leave, and that the synchronizer keeps ordering with one node down but not with two. The extra nodes and apps use small connection pools, so LocalNet stays under its Postgres connection limit. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Found by running the four-node LocalNet and its test locally: - Each BFT orderer gets its own database. Its migrations create a debug schema that the node's storage had already created when both shared one database, so sequencer initialization failed. - The test reaches the sequencers' public APIs through published localhost ports. Onboarding a sequencer calls them from the test, which cannot resolve the canton container's name. - The participant reconnects after switching to node 5 as its only sequencer, as a changed connection only takes effect on reconnect. Also formats the test. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
The startup script skips sync operator users that already exist, so starting the four-node LocalNet again without -D no longer fails. The sequencer threshold proposal now waits a bounded time, so a BFT network that never forms fails the startup instead of hanging it. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
traffic_state_of_members_at_timestamp does not return the state at the given timestamp (see its TODO #25930), so the four-node test reads each node's latest state directly and no longer claims one timestamp. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
96f0fce to
a3e10b8
Compare
Main drops the shared LocalNet test base and the LocalNet upgrade setup, and its sync operator only configures its sequencer. The four-node test now carries its own helpers, as the other LocalNet tests do, and the extra sync operators only point at their own participant and sequencer. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
a3e10b8 to
bd3d2bb
Compare
moritzkiefer-da
left a comment
moritzkiefer-da
left a comment
There was a problem hiding this comment.
please stop putting every test in localnet.
As I mentioned on an earlier PR, localnet is for app devs to have a low resource and easy way to spin up something they can test against. A single node is perfectly sufficient for that, from an app dev's perspective it's irrelveant whether the synchronizer is decentralized.
For us, localnet tests are strictly worse. Instead of just running an integration test you now need to first build the docker images which takes much much longer and running the compose setup then often takes more resources.
I think that was in my mind and actually wanted to ask about it for the 4 node bft testing. As the PR was placed earlier so it added those tests- but Please confirm how we gonna proceed on this-
|
no we don't, localnet should just be a single node.
extend the CI to test with 4 nodes. |
Removes the four-node app-synchronizer profile (-B) and its LocalNet test. The four-node setup moves to a Splice integration test. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Starts a dedicated synchronizer on four BFT nodes with one sync operator per node, and checks that every node grants a traffic purchase and that the nodes agree on what it consumes. Runs in the sync-operator CI job. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
sadiq1971
changed the title
Updated |
|
@sadiq1971 I presume you're aware that your test is failing, and that you're looking into it? |
The test JVM's Canton console asks a sequencer for its protocol version over the public API while generating the genesis state, which an uninitialized sequencer cannot answer. The four-node Canton now runs a bootstrap script instead, as start-canton.sh does with bootstrap-canton.sc. Also matches the other sync operator tests with withOnlyAliceValidatorConnectingToSplitwell, waits for traffic control on every node before the operators start, and formats the test. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Yup. Will ping you once the CI is clean. |
|
@isegall-da Looks good now. |
…embers [ci] Bootstraps the four-node synchronizer at protocol version 36, the version the global synchronizer runs in the tests, and checks that the synchronizer lists all four sequencers and all four mediators with Splice's thresholds. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Traffic control is now on from the bootstrap, as on a dedicated synchronizer. The participant joins by handshake alone, which needs no traffic and makes it a member the operators grant to, then buys traffic and only then connects. The operators drop a purchase for a member the synchronizer does not know yet (ChainSafe/canton-extending-mainnet#116), so the purchase cannot come before the join. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Like the other sync operator suites, the test registers a synchronizer, which needs splice-dso-governance 0.1.30, so the compatibility run against older packages skips it instead of failing it. Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
Signed-off-by: sadiq1971 <sadiqurr8@gmail.com>
3 participants
Part of ChainSafe/canton-extending-mainnet#106
Summary: adds an integration test that runs a dedicated synchronizer on four BFT nodes, one sync operator per node, and checks that all four nodes grant a traffic purchase and agree on what it consumes. LocalNet stays single-node.