This guide provides instructions for upgrading to specific versions of Tendermint Core.
AbciVersion
to RequestInfo
. Applications should check that the ABCI version they expect is being used in order to avoid unimplemented changes errors.SetOption
has been removed from the ABCI.Client interface. This feature was used in the early ABCI implementation's.v1
is enabled, transactions broadcasted via sync
mode may return a successful
response with a transaction hash indicating that the transaction was successfully inserted into
the mempool. While this is true for v0
, the v1
mempool reactor may at a later point in time
evict or even drop this transaction after a hash has been returned. Thus, the user or client must
query for that transaction to check if it is still in the mempool.The configuration file field [fastsync]
has been renamed to [blocksync]
.
The top level configuration file field fast-sync
has moved under the new [blocksync]
field as blocksync.enable
.
blocksync.version = "v1"
and blocksync.version = "v2"
(previously fastsync
)
are no longer supported. Please use v0
instead. During the v0.35 release cycle, v0
was
determined to suit the existing needs and the cost of maintaining the v1
and v2
modules
was determined to be greater than necessary.
All config parameters are now hyphen-case (also known as kebab-case) instead of snake_case. Before restarting the node make sure
you have updated all the variables in your config.toml
file.
Added --mode
flag and mode
config variable on config.toml
for setting Mode of the Node: full
| validator
| seed
(default: full
)
ADR-52
BootstrapPeers
has been added as part of the new p2p stack. This will eventually replace
Seeds
. Bootstrap peers are connected with on startup if needed for peer discovery. Unlike
persistent peers, there's no gaurantee that the node will remain connected with these peers.
configuration values starting with priv-validator-
have moved to the new
priv-validator
section, without the priv-validator-
prefix.
The fast sync process as well as the blockchain package and service has all been renamed to block sync
The format of all tendermint on-disk database keys changes in
0.35. Upgrading nodes must either re-sync all data or run a migration
script provided in this release. The script located in
github.com/tendermint/tendermint/scripts/keymigrate/migrate.go
provides the function Migrate(context.Context, db.DB)
which you can
operationalize as makes sense for your deployment.
For ease of use the tendermint
command includes a CLI version of the
migration script, which you can invoke, as in:
tendermint key-migrate
This reads the configuration file as normal and allows the
--db-backend
and --db-dir
flags to change database operations as
needed.
The migration operation is idempotent and can be run more than once, if needed.
You must now specify the node mode (validator|full|seed) in tendermint init [mode]
The --fast-sync
command line option has been renamed to --blocksync.enable
If you had previously used tendermint gen_node_key
to generate a new node
key, keep in mind that it no longer saves the output to a file. You can use
tendermint init validator
or pipe the output of tendermint gen_node_key
to
$TMHOME/config/node_key.json
:
$ tendermint gen_node_key > $TMHOME/config/node_key.json
CLI commands and flags are all now hyphen-case instead of snake_case. Make sure to adjust any scripts that calls a cli command with snake_casing
The p2p layer was reimplemented as part of the 0.35 release cycle and
all reactors were refactored to accomodate the change. As part of that work these
implementations moved into the internal
package and are no longer
considered part of the public Go API of tendermint. These packages
are:
p2p
mempool
consensus
statesync
blockchain
evidence
Accordingly, the node
package changed to reduce access to
tendermint internals: applications that use tendermint as a library
will need to change to accommodate these changes. Most notably:
The Node
type has become internal, and all constructors return a
service.Service
implementation.
The node.DefaultNewNode
and node.NewNode
constructors are no
longer exported and have been replaced with node.New
and
node.NewDefault
which provide more functional interfaces.
To access any of the functionality previously available via the
node.Node
type, use the *local.Local
"RPC" client, that exposes
the full RPC interface provided as direct function calls. Import the
github.com/tendermint/tendermint/rpc/client/local
package and pass
the node service as in the following:
node := node.NewDefault() //construct the node object
// start and set up the node service
client := local.New(node.(local.NodeService))
// use client object to interact with the node
Mark gRPC in the RPC layer as deprecated and to be removed in 0.36.
When running with the new P2P Layer, the methods UnsafeDialSeeds
and
UnsafeDialPeers
RPC methods will always return an error. They are
deprecated and will be removed in 0.36 when the legacy peer stack is
removed.
Additionally the format of the Peer list returned in the NetInfo
method changes in this release to accommodate the different way that
the new stack tracks data about peers. This change affects users of
both stacks.
The P2P library was reimplemented in this release. The new implementation is enabled by default in this version of Tendermint. The legacy implementation is still included in this version of Tendermint as a backstop to work around unforeseen production issues. The new and legacy version are interoperable. If necessary, you can enable the legacy implementation in the server configuration file.
To make use of the legacy P2P implemementation add or update the following field of
your server's configuration file under the [p2p]
section:
[p2p]
...
use-legacy = true
...
If you need to do this, please consider filing an issue in the Tendermint repository to let us know why. We plan to remove the legacy P2P code in the next (v0.36) release.
The new p2p implementation enables selection of the queue type to be used for passing messages between peers.
The following values may be used when selecting which queue type to use:
fifo
: (default) An unbuffered and lossless queue that passes messages through
in the order in which they were received.
priority
: A priority queue of messages.
wdrr
: A queue implementing the Weighted Deficit Round Robin algorithm. A
weighted deficit round robin queue is created per peer. Each queue contains a
separate 'flow' for each of the channels of communication that exist between any two
peers. Tendermint maintains a channel per message type between peers. Each WDRR
queue maintains a shared buffered with a fixed capacity through which messages on different
flows are passed.
For more information on WDRR scheduling, see: https://en.wikipedia.org/wiki/Deficit_round_robin
To select a queue type, add or update the following field under the [p2p]
section of your server's configuration file.
[p2p]
...
queue-type = wdrr
...
The changes to p2p layer removed existing support for custom reactors. Based on our understanding of how this functionality was used, the introduction of the prioritized mempool covers nearly all of the use cases for custom reactors. If you are currently running custom reactors and mempools and are having trouble seeing the migration path for your project please feel free to reach out to the Tendermint Core development team directly.
Upgrading to Tendermint 0.34 requires a blockchain restart. This release is not compatible with previous blockchains due to changes to the encoding format (see "Protocol Buffers," below) and the block header (see "Blockchain Protocol").
Note also that Tendermint 0.34 also requires Go 1.16 or higher.
The ABCIVersion
is now 0.17.0
.
New ABCI methods (ListSnapshots
, LoadSnapshotChunk
, OfferSnapshot
, and ApplySnapshotChunk
)
were added to support the new State Sync feature.
Previously, syncing a new node to a preexisting network could take days; but with State Sync,
new nodes are able to join a network in a matter of seconds.
Read the spec
if you want to learn more about State Sync, or if you'd like your application to use it.
(If you don't want to support State Sync in your application, you can just implement these new
ABCI methods as no-ops, leaving them empty.)
KV.Pair
has been replaced with abci.EventAttribute
. The EventAttribute.Index
field
allows ABCI applications to dictate which events should be indexed.
The blockchain can now start from an arbitrary initial height,
provided to the application via RequestInitChain.InitialHeight
.
ABCI evidence type is now an enum with two recognized types of evidence:
DUPLICATE_VOTE
and LIGHT_CLIENT_ATTACK
.
Applications should be able to handle these evidence types
(i.e., through slashing or other accountability measures).
The PublicKey
type
(used in ABCI as part of ValidatorUpdate
) now uses a oneof
protobuf type.
Note that since Tendermint only supports ed25519 validator keys, there's only one
option in the oneof
. For more, see "Protocol Buffers," below.
The field Proof
, on the ABCI type ResponseQuery
, is now named ProofOps
.
For more, see "Crypto," below.
The method SetOption
has been removed from the ABCI.Client interface. This feature was used in the early ABCI implementation's.
The default codec is now proto3, not amino. The schema files can be found in the /proto
directory. For more, see "Protobuf," below.
Header#LastResultsHash
, which is the root hash of a Merkle tree built from
ResponseDeliverTx(Code, Data)
as of v0.34 also includes GasWanted
and GasUsed
fields.
Merkle hashes of empty trees previously returned nothing, but now return the hash of an empty input,
to conform with RFC-6962.
This mainly affects Header#DataHash
, Header#LastResultsHash
, and
Header#EvidenceHash
, which are often empty. Non-empty hashes can also be affected, e.g. if their
inputs depend on other (empty) Merkle hashes, giving different results.
Tendermint now relies on the application to tell it which transactions to index. This means that
in the config.toml
, generated by Tendermint, there is no longer a way to specify which
transactions to index. tx.height
and tx.hash
will always be indexed when using the kv
indexer.
Applications must now choose to either a) enable indexing for all transactions, or
b) allow node operators to decide which transactions to index.
Applications can notify Tendermint to index a specific transaction by setting
Index: bool
to true
in the Event Attribute:
[]types.Event{
{
Type: "app",
Attributes: []types.EventAttribute{
{Key: []byte("creator"), Value: []byte("Cosmoshi Netowoko"), Index: true},
},
},
}
Tendermint 0.34 replaces Amino with Protocol Buffers for encoding. This migration is extensive and results in a number of changes, however, Tendermint only uses the types generated from Protocol Buffers for disk and wire serialization. This means that these changes should not affect you as a Tendermint user.
However, Tendermint users and contributors may note the following changes:
/proto
.
This is in line with the recommended file layout by Buf.
For more, see the Buf documentation.PublicKey
type now uses
a oneof
type.For more on the Protobuf changes, please see our blog post on this migration.
Tendermint 0.34 includes new and updated consensus parameters.
AppVersion
, which is the version of the ABCI application.MaxBytes
, which caps the total amount of evidence. The default is 1048576 (1 MB).PubKeyEd25519
to PubKey
. This reduces stutter (e.g., ed25519.PubKey
).[]byte
) instead of byte arrays ([<size>]byte
).cosmos/cosmos-sdk/types/multisig
.merkle
PackageSimpleHashFromMap()
and SimpleProofsFromMap()
were removed.Simple
has been removed. (For example, SimpleProof
is now called Proof
.)/proto
directory.Proof
that contained multiple ProofOp's has been renamed to ProofOps
.
As noted above, this affects the ABCI type ResponseQuery
:
The field that was named Proof is now named ProofOps
.HashFromByteSlices
and ProofsFromByteSlices
now return a hash for empty inputs, to conform with
RFC-6962.libs
PackageThe bech32
package has moved to the Cosmos SDK:
cosmos/cosmos-sdk/types/bech32
.
The tendermint lite
command has been renamed to tendermint light
and has a slightly different API.
See the docs for details.
We have a new, rewritten light client! You can read more about the justifications and details behind this change.
Other user-relevant changes include:
lite
package was removed; the new light client uses the light
package.Verifier
was broken up into two pieces:
VerifyX
functions)Client
object, which represents the complete light clientLightBlock
s/rpc
directory./proxy
directory.state
PackageState.InitialHeight
has been added to record the initial chain height, which must be 1
(not 0
) if starting from height 1
. This can be configured via the genesis field initial_height
.state
package now has a Store
interface. All functions in
state/store.go
are now part of the interface. The interface returns errors on all methods and can be used by calling state.NewStore(dbm.DB)
.privval
PackageAll requests are now accompanied by the chain ID from the network. This is a optional field and can be ignored by key management systems; however, if you are using the same key management system for multiple different blockchains, we recommend that you check the chain ID.
/unsafe_start_cpu_profiler
, /unsafe_stop_cpu_profiler
and
/unsafe_write_heap_profile
were removed.
For profiling, please use the pprof server, which can
be enabled through --rpc.pprof_laddr=X
flag or pprof_laddr=X
config setting
in the rpc section.Content-Type
header returned on RPC calls is now (correctly) set as application/json
.Version is now set through Go linker flags ld_flags
. Applications that are using tendermint as a library should set this at compile time.
Example:
go install -mod=readonly -ldflags "-X github.com/tendermint/tendermint/version.TMCoreSemVer=$(go list -m github.com/tendermint/tendermint | sed 's/ /\@/g') -s -w " -trimpath ./cmd
Additionally, the exported constant version.Version
is now version.TMCoreSemVer
.
rpc/client
HTTP and local clients have been moved into http
and local
subpackages, and their constructors have been renamed to New()
.When upgrading to version 0.33.4 you will have to fetch the third_party
directory along with the updated proto files.
ResponseCommit added a field for block retention. The application can provide information to Tendermint on how to prune blocks.
If an application would like to not prune any blocks pass a 0
in this field.
message ResponseCommit {
// reserve 1
bytes data = 2; // the Merkle root hash
++ uint64 retain_height = 3; // the oldest block height to retain ++
}
This release is not compatible with previous blockchains due to commit becoming signatures only and fields in the header have been removed.
TotalTxs
and NumTxs
were removed from the header. Commit
now consists
mostly of just signatures.
type Commit struct {
Height int64
Round int
BlockID BlockID
Signatures []CommitSig
}
type BlockIDFlag byte
const (
// BlockIDFlagAbsent - no vote was received from a validator.
BlockIDFlagAbsent BlockIDFlag = 0x01
// BlockIDFlagCommit - voted for the Commit.BlockID.
BlockIDFlagCommit = 0x02
// BlockIDFlagNil - voted for nil.
BlockIDFlagNil = 0x03
)
type CommitSig struct {
BlockIDFlag BlockIDFlag
ValidatorAddress Address
Timestamp time.Time
Signature []byte
}
See #63 for the complete spec change.
The secret connection now includes a transcript hashing. If you want to implement a handshake (or otherwise have an existing implementation), you'll need to make the same changes that were made here.
You will need to generate a new config if you have used a prior version of tendermint.
Tags have been entirely renamed throughout the codebase to events and there keys are called compositeKeys.
Evidence Params has been changed to include duration.
consensus_params.evidence.max_age_duration
.consensus_params.evidence.max_age
to max_age_num_blocks
.libs/common
has been removed in favor of specific pkgs.
async
service
rand
net
strings
cmap
errors
pkg/validators
is now paginated (default: 30 vals per page)/block_results
response format updated see RPC docs for detailsjson-client-XYZ
This release is compatible with previous blockchains, however the new ABCI Events mechanism may create some complexity for nodes wishing to continue operation with v0.32 from a previous version. There are some minor breaking changes to the RPC.
If you have db_backend
set to leveldb
in your config file, please change it
to goleveldb
or cleveldb
.
The default listen address for the RPC is now 127.0.0.1
. If you want to expose
it publicly, you have to explicitly configure it. Note exposing the RPC to the
public internet may not be safe - endpoints which return a lot of data may
enable resource exhaustion attacks on your node, causing the process to crash.
Any consumers of /block_results
need to be mindful of the change in all field
names from CamelCase to Snake case, eg. results.DeliverTx
is now results.deliver_tx
.
This is a fix, but it's breaking.
ABCI responses which previously had a Tags
field now have an Events
field
instead. The original Tags
field was simply a list of key-value pairs, where
each key effectively represented some attribute of an event occuring in the
blockchain, like sender
, receiver
, or amount
. However, it was difficult to
represent the occurence of multiple events (for instance, multiple transfers) in a single list.
The new Events
field contains a list of Event
, where each Event
is itself a list
of key-value pairs, allowing for more natural expression of multiple events in
eg. a single DeliverTx or EndBlock. Note each Event
also includes a Type
, which is meant to categorize the
event.
For transaction indexing, the index key is
prefixed with the event type: {eventType}.{attributeKey}
.
If the same event type and attribute key appear multiple times, the values are
appended in a list.
To make queries, include the event type as a prefix. For instance if you
previously queried for recipient = 'XYZ'
, and after the upgrade you name your event transfer
,
the new query would be for transfer.recipient = 'XYZ'
.
Note that transactions indexed on a node before upgrading to v0.32 will still be indexed
using the old scheme. For instance, if a node upgraded at height 100,
transactions before 100 would be queried with recipient = 'XYZ'
and
transactions after 100 would be queried with transfer.recipient = 'XYZ'
.
While this presents additional complexity to clients, it avoids the need to
reindex. Of course, you can reset the node and sync from scratch to re-index
entirely using the new scheme.
We illustrate further with a more complete example.
Prior to the update, suppose your ResponseDeliverTx
look like:
abci.ResponseDeliverTx{
Tags: []kv.Pair{
{Key: []byte("sender"), Value: []byte("foo")},
{Key: []byte("recipient"), Value: []byte("bar")},
{Key: []byte("amount"), Value: []byte("35")},
}
}
The following queries would match this transaction:
query.MustParse("tm.event = 'Tx' AND sender = 'foo'")
query.MustParse("tm.event = 'Tx' AND recipient = 'bar'")
query.MustParse("tm.event = 'Tx' AND sender = 'foo' AND recipient = 'bar'")
Following the upgrade, your ResponseDeliverTx
would look something like:
the following Events
:
abci.ResponseDeliverTx{
Events: []abci.Event{
{
Type: "transfer",
Attributes: kv.Pairs{
{Key: []byte("sender"), Value: []byte("foo")},
{Key: []byte("recipient"), Value: []byte("bar")},
{Key: []byte("amount"), Value: []byte("35")},
},
}
}
Now the following queries would match this transaction:
query.MustParse("tm.event = 'Tx' AND transfer.sender = 'foo'")
query.MustParse("tm.event = 'Tx' AND transfer.recipient = 'bar'")
query.MustParse("tm.event = 'Tx' AND transfer.sender = 'foo' AND transfer.recipient = 'bar'")
For further documentation on Events
, see the docs.
The ABCI Application interface changed slightly so the CheckTx and DeliverTx methods now take Request structs. The contents of these structs are just the raw tx bytes, which were previously passed in as the argument.
There are no breaking changes in this release except Go API of p2p and mempool packages. Hovewer, if you're using cleveldb, you'll need to change the compilation tag:
Use cleveldb
tag instead of gcc
to compile Tendermint with CLevelDB or
use make build_c
/ make install_c
(full instructions can be found at
https://tendermint.com/docs/introduction/install.html#compile-with-cleveldb-support)
This release contains a breaking change to the behaviour of the pubsub system. It also contains some minor breaking changes in the Go API and ABCI. There are no changes to the block or p2p protocols, so v0.31.0 should work fine with blockchains created from the v0.30 series.
The pubsub no longer blocks on publishing. This may cause some WebSocket (WS) clients to stop working as expected. If your WS client is not consuming events fast enough, Tendermint can terminate the subscription. In this case, the WS client will receive an error with description:
{
"jsonrpc": "2.0",
"id": "{ID}#event",
"error": {
"code": -32000,
"msg": "Server error",
"data": "subscription was canceled (reason: client is not pulling messages fast enough)" // or "subscription was canceled (reason: Tendermint exited)"
}
}
Additionally, there are now limits on the number of subscribers and
subscriptions that can be active at once. See the new
`rpc.max_subscription_clients` and `rpc.max_subscriptions_per_client` values to
configure this.
Simple rename of ConsensusParams.BlockSize
to ConsensusParams.Block
.
The ConsensusParams.Block.TimeIotaMS
field was also removed. It's configured
in the ConsensusParsm in genesis.
See the CHANGELOG. These are relatively straight forward.
This release contains a breaking change to both the block and p2p protocols, however it may be compatible with blockchains created with v0.29.0 depending on the chain history. If your blockchain has not included any pieces of evidence, or no piece of evidence has been included in more than one block, and if your application has never returned multiple updates for the same validator in a single block, then v0.30.0 will work fine with blockchains created with v0.29.0.
The p2p protocol change is to fix the proposer selection algorithm again. Note that proposer selection is purely a p2p concern right now since the algorithm is only relevant during real time consensus. This change is thus compatible with v0.29.0, but all nodes must be upgraded to avoid disagreements on the proposer.
Applications must ensure they do not return duplicates in
ResponseEndBlock.ValidatorUpdates
. A pubkey must only appear once per set of
updates. Duplicates will cause irrecoverable failure. If you have a very good
reason why we shouldn't do this, please open an issue.
This release contains some breaking changes to the block and p2p protocols, and will not be compatible with any previous versions of the software, primarily due to changes in how various data structures are hashed.
Any implementations of Tendermint blockchain verification, including lite clients, will need to be updated. For specific details:
There was also a small change to field ordering in the vote struct. Any implementations of an out-of-process validator (like a Key-Management Server) will need to be updated. For specific details:
Finally, the proposer selection algorithm continues to evolve. See the work-in-progress specification.
For everything else, please see the CHANGELOG.
This release breaks the format for the priv_validator.json
file
and the protocol used for the external validator process.
It is compatible with v0.27.0 blockchains (neither the BlockProtocol nor the
P2PProtocol have changed).
Please read carefully for details about upgrading.
Note: Backup your config/priv_validator.json
before proceeding.
priv_validator.json
The config/priv_validator.json
is now two files:
config/priv_validator_key.json
and data/priv_validator_state.json
.
The former contains the key material, the later contains the details on the last
message signed.
When running v0.28.0 for the first time, it will back up any pre-existing
priv_validator.json
file and proceed to split it into the two new files.
Upgrading should happen automatically without problem.
To upgrade manually, use the provided privValUpgrade.go
script, with exact paths for the old
priv_validator.json
and the locations for the two new files. It's recomended
to use the default paths, of config/priv_validator_key.json
and
data/priv_validator_state.json
, respectively:
go run scripts/privValUpgrade.go <old-path> <new-key-path> <new-state-path>
The Unix and TCP implementations of the remote signing validator have been consolidated into a single implementation. Thus in both cases, the external process is expected to dial Tendermint. This is different from how Unix sockets used to work, where Tendermint dialed the external process.
The PubKeyMsg
was also split into separate Request
and Response
types
for consistency with other messages.
Note that the TCP sockets don't yet use a persistent key, so while they're encrypted, they can't yet be properly authenticated. See #3105. Note the Unix socket has neither encryption nor authentication, but will add a shared-secret in #3099.
This release contains some breaking changes to the block and p2p protocols, but does not change any core data structures, so it should be compatible with existing blockchains from the v0.26 series that only used Ed25519 validator keys. Blockchains using Secp256k1 for validators will not be compatible. This is due to the fact that we now enforce which key types validators can use as a consensus param. The default is Ed25519, and Secp256k1 must be activated explicitly.
It is recommended to upgrade all nodes at once to avoid incompatibilities at the
peer layer - namely, the heartbeat consensus message has been removed (only
relevant if create_empty_blocks=false
or create_empty_blocks_interval > 0
),
and the proposer selection algorithm has changed. Since proposer information is
never included in the blockchain, this change only affects the peer layer.
The ReverseIterator API has changed the meaning of start
and end
.
Before, iteration was from start
to end
, where
start > end
. Now, iteration is from end
to start
, where start < end
.
The iterator also excludes end
. This change allows a simplified and more
intuitive logic, aligning the semantic meaning of start
and end
in the
Iterator
and ReverseIterator
.
This release enforces a new consensus parameter, the ValidatorParams.PubKeyTypes. Applications must ensure that they only return validator updates with the allowed PubKeyTypes. If a validator update includes a pubkey type that is not included in the ConsensusParams.Validator.PubKeyTypes, block execution will fail and the consensus will halt.
By default, only Ed25519 pubkeys may be used for validators. Enabling Secp256k1 requires explicit modification of the ConsensusParams. Please update your application accordingly (ie. restrict validators to only be able to use Ed25519 keys, or explicitly add additional key types to the genesis file).
This release contains a lot of changes to core data types and protocols. It is not compatible to the old versions and there is no straight forward way to update old data to be compatible with the new version.
To reset the state do:
tendermint unsafe_reset_all
Here we summarize some other notable changes to be mindful of.
All timeouts must be changed from integers to strings with their duration, for
instance flush_throttle_timeout = 100
would be changed to
flush_throttle_timeout = "100ms"
and timeout_propose = 3000
would be changed
to timeout_propose = "3s"
.
The default behaviour of /abci_query
has been changed to not return a proof,
and the name of the parameter that controls this has been changed from trusted
to prove
. To get proofs with your queries, ensure you set prove=true
.
Various version fields like amino_version
, p2p_version
, consensus_version
,
and rpc_version
have been removed from the node_info.other
and are
consolidated under the tendermint semantic version (ie. node_info.version
) and
the new block
and p2p
protocol versions under node_info.protocol_version
.
Field numbers were bumped in the Header
and ResponseInfo
messages to make
room for new version
fields. It should be straight forward to recompile the
protobuf file for these changes.
The ResponseQuery.Proof
field is now structured as a []ProofOp
to support
generalized Merkle tree constructions where the leaves of one Merkle tree are
the root of another. If you don't need this functionality, and you used to
return <proof bytes>
here, you should instead return a single ProofOp
with
just the Data
field set:
[]ProofOp{
ProofOp{
Data: <proof bytes>,
}
}
For more information, see:
The merkle.Hasher
interface was removed. Functions which used to take Hasher
now simply take []byte
. This means that any objects being Merklized should be
serialized before they are passed in.
The node.RunForever
function was removed. Signal handling and running forever
should instead be explicitly configured by the caller. See how we do it
here.
All hashes, except for public key addresses, are now 32-bytes.
This release has minimal impact.
If you use GasWanted in ABCI and want to enforce it, set the MaxGas in the genesis file (default is no max).
New 0.24.0 release contains a lot of changes to the state and types. It's not compatible to the old versions and there is no straight forward way to update old data to be compatible with the new version.
To reset the state do:
tendermint unsafe_reset_all
Here we summarize some other notable changes to be mindful of.
p2p.max_num_peers
was removed in favor of p2p.max_num_inbound_peers
and
p2p.max_num_outbound_peers
.
# Maximum number of inbound peers
max_num_inbound_peers = 40
# Maximum number of outbound peers to connect to, excluding persistent peers
max_num_outbound_peers = 10
As you can see, the default ratio of inbound/outbound peers is 4/1. The reason is we want it to be easier for new nodes to connect to the network. You can tweak these parameters to alter the network topology.
The result of /commit
used to contain header
and commit
fields at the top level. These are now contained under the signed_header
field.
The header has been upgraded and contains new fields, but none of the existing fields were changed, except their order.
The Validator
type was split into two, one containing an Address
and one
containing a PubKey
. When processing RequestBeginBlock
, use the Validator
type, which contains just the Address
. When returning ResponseEndBlock
, use
the ValidatorUpdate
type, which contains just the PubKey
.
Validator set updates returned in ResponseEndBlock for height H
used to take
effect immediately at height H+1
. Now they will be delayed one block, to take
effect at height H+2
. Note this means that the change will be seen by the ABCI
app in the RequestBeginBlock.LastCommitInfo
at block H+3
. Apps were already
required to maintain a map from validator addresses to pubkeys since v0.23 (when
pubkeys were removed from RequestBeginBlock), but now they may need to track
multiple validator sets at once to accomodate this delay.
The ConsensusParams.BlockSize.MaxTxs
was removed in favour of
ConsensusParams.BlockSize.MaxBytes
, which is now enforced. This means blocks
are limitted only by byte-size, not by number of transactions.