* p2p: panic on transport error
Addresses #2823. Currently, the acceptRoutine exits if the transport returns
an error trying to accept a new connection. Once this happens, the node
can't accept any new connections. So here, we panic instead. While we
could potentially be more intelligent by rerunning the acceptRoutine, the
error may indicate something more fundamental (eg. file desriptor limit)
that requires a restart anyways. We can leave it to process managers to
handle that restart, and notify operators about the panic.
* changelog
deliverTxResCh, like any other eventBus (pubsub) channel, is closed when
eventBus is stopped. We must check if the channel is still open. The
alternative approach is to not close any channels, which seems a bit
odd.
Fixes#2408
* WIP: tests for #2785
* rebase onto develop
* add Bucky's test without changing ValidatorSet.Update
* make TestValidatorSetBasic fail
* add ProposerPriority preserving fix to ValidatorSet.Update to fix
TestValidatorSetBasic
* fix randValidator_ to stay in bounds of MaxTotalVotingPower
* check for expected proposer and remove some duplicate code
* actually limit the voting power of random validator ...
* fix test
* types: ValidatorSet.Update preserves ProposerPriority
This solves the other issue discovered as part of #2718,
where Accum (now called ProposerPriority) is reset to
0 every time a validator is updated.
* update changelog
* add test
* update comment
* Update types/validator_set_test.go
Co-Authored-By: ebuchman <ethan@coinculture.info>
* set the accum of a new validator to (-total voting power):
- disincentivize validators to unbond, then rebon to reset their
negative Accum to zero
additional unrelated changes:
- do not capitalize error msgs
- fix typo
* review comments: (re)capitalize errors & delete obsolete comments
* More changes suggested by @melekes
* WIP: do not batch clip (#2809)
* substract avgAccum on each iteration
- temporarily skip test
* remove unused method safeMulClip / safeMul
* always substract the avg accum
- temp. skip another test
* remove overflow / underflow tests & add tests for avgAccum:
- add test for computeAvgAccum
- as we substract the avgAccum now we will not trivially over/underflow
* address @cwgoes' comments
* shift by avg at the end of IncrementAccum
* Add comment to MaxTotalVotingPower
* Guard inputs to not exceed MaxTotalVotingPower
* Address review comments:
- do not fetch current validator from set again
- update error message
* Address a few review comments:
- fix typo
- extract variable
* address more review comments:
- clarify 1.125*totalVotingPower == totalVotingPower + (totalVotingPower >> 3)
* review comments: panic instead of "clipping":
- total voting power is guarded to not exceed MaxTotalVotingPower ->
panic if this invariant is violated
* fix failing test
* Enforce validators can only use the correct pubkey type
* adapt to variable renames
* Address comments from #2636
* separate updating and validation logic
* update spec
* Add test case for TestStringSliceEqual, clarify slice copying code
* Address @ebuchman's comments
* Split up testing validator update execution, and its validation
* update docs
- make install_c cmd (install)
- explain node IDs (quick-start)
- update UPGRADING section (using-tendermint)
* use git clone with JS example
JS devs may not have Go installed and we should not force them to.
* rewrite sentence
Modify lookForHeight to return a height only there's a equal operator.
Previously, it was returning a height even for range conditions: "height
< 10000".
Fixes#2759
* Fixed accepting integer IDs in requests for Tendermint RPC server (#2366)
* added a wrapper interface `jsonrpcid` that represents both string and int IDs in JSON-RPC requests/responses + custom JSON unmarshallers
* changed client-side code in RPC that uses it
* added extra tests for integer IDs
* updated CHANGELOG_PENDING, as suggested by PR instructions
* addressed PR comments
* added table driven tests for request type marshalling/unmarshalling
* expanded handler test to check IDs
* changed pending changelog note
* changed json rpc request/response unmarshalling to use empty interfaces and type switches on ID
* some cleanup
Within every tx in the mempool, we store a 64 bit counter,
as an index for when it was inserted into the mempool.
This counter doesn't really serve any purpose.
It was likely added for debugging at one point,
Removing the counter reclaims memory,
which enables greater mempool sizes / mitigates resources at the same size.
Closes#2835
* #2815 do not broadcast heartbeat proposal when we are non-validator
* #2815 adding preliminary changelog entry
* #2815 cosmetics and added test
* #2815 missed a little detail
- it's enough to call getAddress() once here
* #2815 remove debug logging from tests
* #2815 OK. I seem to be doing something fundamentally wrong here
* #2815 next iteration of proposalHeartbeat tests
- try and use "ensure" pattern in common_test
* 2815 incorporate review comments
* Replaces our current http servers where connections stay open forever with ones with timeouts to prevent file descriptor exhaustion
* Use the correct handler
* Put in go routines
* fix err
* changelog
* rpc: export Read/WriteTimeout
The `broadcast_tx_commit` endpoint has it's own timeout.
If this is longer than the http server's WriteTimeout, the
user will receive an error. Here, we export the WriteTimeout
and set the broadcast_tx_commit timeout to be less than it.
In the future, we should use a config struct for the timeouts
to avoid the need to export. The broadcast_tx_commit timeout
may also become configurable, but we must check that it's less
than the server's WriteTimeout.
* p2p/conn: FlushStop. Use in pex. Closes#2092
In seed mode, we call StopPeer immediately after Send.
Since flushing msgs to the peer happens in the background,
the peer connection is often closed before the messages are
actually sent out. The new FlushStop method allows all msgs
to first be written and flushed out on the conn before it is closed.
* fix dummy peer
* typo
* fixes from review
* more comments
* ensure pex doesn't call FlushStop more than once
FlushStop is not safe to call more than once,
but we call it from Receive in a go-routine so Receive
doesn't block.
To ensure we only call it once, we use the lastReceivedRequests
map - if an entry already exists, then FlushStop should already have
been called and we can return.
Ethan Buchman
Anton Kaliaev
Ismail Khoffi
Dev Ojha
Daniil Lashin
Zach
srmo
Jae Kwon
nagarajmanjunath
Jernej Kos
JamesRay
Tomas Tauber
Joe Bowman
cong
Zaki Manian