Client: go
Provide ExtractIDLExceptionClientMiddleware client middleware
implementation and ExtractExceptionFromResult to extract exceptions
defined in thrift IDL into err return so they are accessible from other
client middlewares.
Client: go
Currently go's TDebugProtocol does two things: log the read/writes, and
duplicate all the reads/writes to another TProtocol. For people who only
need the second feature, even if they use NopLogger for the logging
part, they still need to pay the price of all the fmt.Sprintf calls used
by logging, and on some cases those fmt.Sprintf calls alone can cause
significant CPU to be wasted (in one of our services fmt.Sprintf called
by TDebugProtocol used ~10% of CPU).
Create a dedicated TDuplicateToProtocol to reduce cpu waste, and mark
TDebugProtocol.DuplicateTo as deprecated.
Client: go
Currently in the compiler generated go code, adding a new exception to
an existing endpoint can cause unexpected behaviors when the client
isn't updated. Fix the issue.
Will be cherry-picked into 0.15.0 after merged.
Client: go
Currently in the compiler generated go code, adding a new exception to
an existing endpoint can cause unexpected behaviors when the client
isn't updated. Fix the issue.
Will be cherry-picked into 0.15.0 after merged.
Client: go
Staticcheck is the recommended replacement of the frozen and deprecated
official golint linter [1].
Fix the things it complained about (or add lint:ignore directive) in:
- lib/go/thrift
- lib/go/test/tests
- tutorial/go/src
- test/go/src
- compiler generated code
The majority of the fixes are in the following categories:
- Use of deprecated function (mainly the TConfiguration related ones)
- Redundant break in switch cases
- Unused and unexported variables/fields/functions
Also in the same spirit as fb539ae, remove the error return from
NewTSSLSocket as it can never be non-nil.
This change will be cherry-picked into 0.15.0 branch after merged.
[1]: https://groups.google.com/g/golang-nuts/c/rCP70Aq_tBc
Client: go
Staticcheck is the recommended replacement of the frozen and deprecated
official golint linter [1].
Fix the things it complained about (or add lint:ignore directive) in:
- lib/go/thrift
- lib/go/test/tests
- tutorial/go/src
- test/go/src
- compiler generated code
The majority of the fixes are in the following categories:
- Use of deprecated function (mainly the TConfiguration related ones)
- Redundant break in switch cases
- Unused and unexported variables/fields/functions
Also in the same spirit as fb539ae, remove the error return from
NewTSSLSocket as it can never be non-nil.
This change will be cherry-picked into 0.15.0 branch after merged.
[1]: https://groups.google.com/g/golang-nuts/c/rCP70Aq_tBc
Client: go
We used to do DNS lookups in NewTSocketConf, without any timeout checks.
Stop doing that and do DNS lookups in TSocket.Open instead, which
already checks for ConnectTimeout set in TConfiguration.
Also remove the error return from NewTSocketConf.
Client: go
We used to do DNS lookups in NewTSocketConf, without any timeout checks.
Stop doing that and do DNS lookups in TSocket.Open instead, which
already checks for ConnectTimeout set in TConfiguration.
Also remove the error return from NewTSocketConf.
This change improves two problems in go code imports:
1. Always rename import the thrift package into "thrift", as we allow
the user to use a different library to replace the official one from
the compiler command line, this makes sure that in compiler generated
go code we can always blindly use "thrift.*".
2. We added auto rename import dedup in d9019fc5a4, but in that change
for system packages we always use the full import path as the dedup
identifier, so system package "database/sql/driver" would not be
detected as a conflict against a thrift go namespace of
"foo.bar.driver". Use the part after the last "/" in system packages
as the dedup identifier instead.
Client: go
Currently we only treat TTransportException with typeId == TIMED_OUT as
timeout (return true in Timeout function). When opening a new socket, if
we got a connect timeout from net.Dial, we wrap the error as
TTransportException with typeId == NOT_OPEN, thus it's no longer treated
as a timeout error.
Change the error to be directly wrapping the original error (instead of
recreate a new error with the same error message), and change
tTransportException.Timeout to also return true if the wrapped error
is a timeout error. This way we don't have to break anything (if code
rely on TTransportException.TypeId being NOT_OPEN in this case, that's
still true).
While I'm here, also update CHANGES.md from #2359.
Client: go
Make a breaking change so that TClient.Call returns the response
meta, currently only contains headers but could be expanded in the
future, and make a compiler change to compiler generated clients to take
advantage of that and provide access to response metadata to users.
Client: go
Add TExceptionType enum type, and add
TExceptionType() TExceptionType
function to TException definition.
Also make TProtocolException unwrap-able.
Client: go
Define TConfiguration following the spec, and also move the following
configurations scattered around different TTransport/TProtocol into it:
- connect and socket timeouts for TSocket and TSSLSocket
- tls config for TSSLSocket
- max frame size for TFramedTransport
- strict read and strict write for TBinaryProtocol
- proto id for THeaderTransport
Also add TConfiguration support for the following and their factories:
- THeaderTransport and THeaderProtocol
- TBinaryProtocol
- TCompactProtocol
- TFramedTransport
- TSocket
- TSSLSocket
Also define TConfigurationSetter interface for easier TConfiguration
propagation between wrapped TTransports/TProtocols , and add
implementations to the following for propagation
(they don't use anything from TConfiguration themselves):
- StreamTransport
- TBufferedTransport
- TDebugProtocol
- TJSONProtocol
- TSimpleJSONProtocol
- TZlibTransport
TConfigurationSetter are not implemented by the factories of the
"propagation only" TTransports/TProtocols, if they have a factory. For
those use cases, TTransportFactoryConf and TProtocolFactoryConf are
provided to wrap a factory with the ability to propagate TConfiguration.
Also add simple sanity check for TBinaryProtocol and TCompactProtocol's
ReadString and ReadBinary functions. Currently it only report error if
the header length is larger than MaxMessageSize configured in
TConfiguration, for simplicity.
Client: go
In compiler generated TProcessorFunction implementations, add a
goroutine after read the request to do connectivity check on the input
transport. If the transport is no longer open, cancel the context object
passed into the handler implementation.
Also define ErrAbandonRequest error, to help TSimpleServer closing
client connections that's already closed on the other end.
Client: go
We separated timeout in go's TSocket into connect timeout and socket
timeout in 81334cd, this change does the same for TSSLSocket to keep
them consistent.
Also rename the arg in NewTSocketFromConnTimeout from connTimeout to
socketTimeout, because in that function we already have a connection,
so connect timeout is never used again. The timeout passed into that
function is really for socket timeout, not connect timeout. The name of
that function actually means "New TSocket From Conn (with) Timeout", not
"New TSocket From ConnTimeout" (I guess that's where the original
confusion came from).
Also add the missing change note for the breaking change.
Client: go
As discussed in the JIRA ticket, this commit changes how we handle I/O
timeouts in the go library.
This is a breaking change that adds context to all Read*, Write*, and
Skip functions to TProtocol, along with the compiler change to support
that, and also adds context to TStandardClient.Recv, TDeserializer,
TStruct, and a few others.
Along with the function signature changes, this commit also implements
context cancellation check in the following TProtocol's ReadMessageBegin
implementations:
- TBinaryProtocol
- TCompactProtocol
- THeaderProtocol
In those ReadMessageBegin implementations, if the passed in context
object has a deadline attached, it will keep retrying the I/O timeout
errors, until the deadline on the context object passed. They won't
retry I/O timeout errors if the passed in context does not have a
deadline attached (still return on the first error).
This commit adds a simple middleware framework for Go servers.
It provides:
* A `ProcessorMiddleware` function interface used to define the actual middleware
* `WrapProcessor`, the function that you use to wrap a `TProcessor` in a list of middleware
* A helper `WrappedTProcessorFunction` struct to help with developing middleware
This is a breaking change for any custom implementations of the `TProcessor`
interface, but does not effect the code generated by compiling Thrift files. It
adds two functions to the interface that are a part of the generated `TProcessor`
code, but were not defined in the interface explicitly.
Client: go
This change improves performance when using TDeserializer with a
resource pool. See https://issues.apache.org/jira/browse/THRIFT-5069 for
more context.
Also add TSerializerPool and TDeserializerPool, which are thread-safe
versions of TSerializer and TDeserializer. Benchmark result shows that
they are both faster and use less memory than the plain version:
$ go test -bench Serializer -benchmem
goos: darwin
goarch: amd64
BenchmarkSerializer/baseline-8 577558 1930 ns/op 512 B/op 6 allocs/op
BenchmarkSerializer/plain-8 452712 2638 ns/op 2976 B/op 16 allocs/op
BenchmarkSerializer/pool-8 591698 2032 ns/op 512 B/op 6 allocs/op
PASS
