I observed a couple of problems with the generator in some recent tests:
- there were a couple of hybrid test cases which did not have any
legacy nodes (randomness and all.) I change the probability to
produce more reliable results.
- added options to the generation to be able to add a max (to
compliment the earlier min) number of nodes for local testing.
- added an option to support reversing the sort order so "more
complex" networks were first, as well as tweaked some of the point
values.
- this refactored the generators cli parsing to be a bit more clear.
The main effect of this change is to flush the socket client and server message
encoding buffers immediately once the message is fully and correctly encoded.
This allows us to remove the timer and some other special cases, without
changing the observed behaviour of the system.
-- Background
The socket protocol client and server each use a buffered writer to encode
request and response messages onto the underlying connection. This reduces the
possibility of a single message being split across multiple writes, but has the
side-effect that a request may remain buffered for some time.
The implementation worked around this by keeping a ticker that occasionally
triggers a flush, and by flushing the writer in response to an explicit request
baked into the client/server protocol (see also #6994).
These workarounds are both unnecessary: Once a message has been dequeued for
sending and fully encoded in wire format, there is no real use keeping all or
part of it buffered locally. Moreover, using an asynchronous process to flush
the buffer makes the round-trip performance of the request unpredictable.
-- Benchmarks
Code: https://play.golang.org/p/0ChUOxJOiHt
I found no pre-existing performance benchmarks to justify the flush pattern,
but a natural question is whether this will significantly harm client/server
performance. To test this, I implemented a simple benchmark that transfers
randomly-sized byte buffers from a no-op "client" to a no-op "server" over a
Unix-domain socket, using a buffered writer, both with and without explicit
flushes after each write.
As the following data show, flushing every time (FLUSH=true) does reduce raw
throughput, but not by a significant amount except for very small request
sizes, where the transfer time is already trivial (1.9μs). Given that the
client is calibrated for 1MiB transactions, the overhead is not meaningful.
The percentage in each section is the speedup for flushing only when the buffer
is full, relative to flushing every block. The benchmark uses the default
buffer size (4096 bytes), which is the same value used by the socket client and
server implementation:
FLUSH NBLOCKS MAX AVG TOTAL ELAPSED TIME/BLOCK
false 3957471 512 255 1011165416 2.00018873s 505ns
true 1068568 512 255 273064368 2.000217051s 1.871µs
(73%)
false 536096 4096 2048 1098066401 2.000229108s 3.731µs
true 477911 4096 2047 978746731 2.000177825s 4.185µs
(10.8%)
false 124595 16384 8181 1019340160 2.000235086s 16.053µs
true 120995 16384 8179 989703064 2.000329349s 16.532µs
(2.9%)
false 2114 1048576 525693 1111316541 2.000479928s 946.3µs
true 2083 1048576 526379 1096449173 2.001817137s 961.025µs
(1.5%)
Note also that the FLUSH=false baseline is actually faster than the production
code, which flushes more often than is required by the buffer filling up.
Moreover, the timer slows down the overall transaction rate of the client and
server, indepenedent of how fast the socket transfer is, so the loss on a real
workload is probably much less.
I've been noticing that there are a number of situations where the
statesync reactor blocks waiting for peers (or similar,) I've moved
things around to improve outcomes in local tests.
In the last run, there were two problems at the RPC layer returned
from light nodes' RPC end points. I think exercising the light client
proxy RPC system is something that can/should be done via unit
testing, and that likely these errors are (in production) transient
and (in CI) very likely to fail for test environment issues.
The code in the Tendermint repository makes heavy use of import aliasing.
This is made necessary by our extensive reuse of common base package names, and
by repetition of similar names across different subdirectories.
Unfortunately we have not been very consistent about which packages we alias in
various circumstances, and the aliases we use vary. In the spirit of the advice
in the style guide and https://github.com/golang/go/wiki/CodeReviewComments#imports,
his change makes an effort to clean up and normalize import aliasing.
This change makes no API or behavioral changes. It is a pure cleanup intended
o help make the code more readable to developers (including myself) trying to
understand what is being imported where.
Only unexported names have been modified, and the changes were generated and
applied mechanically with gofmt -r and comby, respecting the lexical and
syntactic rules of Go. Even so, I did not fix every inconsistency. Where the
changes would be too disruptive, I left it alone.
The principles I followed in this cleanup are:
- Remove aliases that restate the package name.
- Remove aliases where the base package name is unambiguous.
- Move overly-terse abbreviations from the import to the usage site.
- Fix lexical issues (remove underscores, remove capitalization).
- Fix import groupings to more closely match the style guide.
- Group blank (side-effecting) imports and ensure they are commented.
- Add aliases to multiple imports with the same base package name.
We moved some files further down in the directory structure in #6964, which
caused the relative paths to the mockery wrapper to stop working.
There does not seem to be an obvious way to get the module root as a default
environment variable, so for now I just added the extra up-slashes.
This package is not used in the tendermint repository since 31e7cdee.
Note that this is not the same package as rpc/client/mock (N.B. singular) which
is still used in some tests.
A search of GitHub turns up only 11 uses, all of which are in clones of the
tendermint repo at old commits..
* rpc: Strip down the base RPC client interface.
Prior to this change, the RPC client interface requires implementing the entire
Service interface, but most of the methods of Service are not needed by the
concrete clients. Dissociate the Client interface from the Service interface.
- Extract only those methods of Service that are necessary to make the existing
clients work.
- Update the clients to combine Start/Onstart and Stop/OnStop. This does not
change what the clients do to start or stop. Only the websocket clients make
use of this functionality anyway.
The websocket implementation uses some plumbing from the BaseService helper.
We should be able to excising that entirely, but the current interface
dependencies among the clients would require a much larger change, and one
that leaks into other (non-RPC) packages.
As a less-invasive intermediate step, preserve the existing client behaviour
(and tests) by extracting the necessary subset of the BaseService
functionality to an analogous RunState helper for clients. I plan to obsolete
that type in a future PR, but for now this makes a useful waypoint.
Related:
- Clean up client implementations.
- Update mocks.
## Description
- Add deprecated to config values in toml
- update config in configuration doc
- explain how to set up a node with the new network
- add sentence about not needing to fork tendermint for built-in tutorial
- closes#6865
- add note to use a released version of tendermint with the tutorials. This is to avoid unknown issues prior to a release.
If the e2e tests error, they leave all of the e2e state around including containers and networks etc.
We should clean this up when the tests shuts down, even if it exits in error.
This should address last night's failure. We've taken the perspective
of "the load generator shouldn't cause tests to fail" in recent
days/weeks, and I think this is just a next step along that line. The
e2e tests shouldn't test performance.
I included some comments indicating the ways that this isn't ideal (it
is perhaps not), and I think that if test networks could make
assertions about the required rate, that might be a cool future
improvement (and good, perhaps, for system benchmarking.)
This document attempts to capture and discuss some of the areas of Tendermint that seem to be cited as causing performance issue. I'm hoping to continue to gather feedback and input on this document to better understand what issues Tendermint performance may cause for our users.
The overall goal of this document is to allow the maintainers and community to get a better sense of these issues and to be more capably able to discuss them and weight trade-offs about any proposed performance-focused changes. This document does not aim to propose any performance improvements. It does suggest useful places for benchmarks and places where additional metrics would be useful for diagnosing and further understanding Tendermint performance.
Please comment with areas where my reasoning seems off or with additional areas that Tendermint performance may be causing user pain.
Sam Kleinman
lklimek
dependabot[bot]
Callum Waters
M. J. Fromberger
Marko
Daria K
JayT106
Ismail Khoffi
William Banfield