Summary: Wasn't causing any issues because it's a noop with the binary protocol
Reviewed By: jwang
Test Plan: honky tonk, no diff in functionality
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665140 13f79535-47bb-0310-9956-ffa450edef68
Summary: New Thrift TBinaryProtocol with a version identifier
Reviewed By: aditya, eugene
Test Plan: Modify your services to have strictRead_ and strictWrite_ both set to FALSE. Then redeploy your services and test running clients against them. Once you have clients and servers running stably on this new code, you should redploy versions with strictWrite_ set to TRUE. Once that's all good, we can set strictRead_ to TRUE as well, and eventually deprecate the old protocol code entirely.
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665138 13f79535-47bb-0310-9956-ffa450edef68
Summary: If enabled, the callback function will be run after the listening socket is opened and the event loop initialized, just prior to calling event_loop()
This is handy if you want to defer some of your initialization until after the socket is open -- allowing clients to connect, but not processing requests until the initialization is complete. I use this in the Synapse tablet server to minimize the communications interruption that happens during a tablet split (replaying commit logs after opening the listening socket).
Review: mcslee
Test Plan: compiled it into Synapse's tablet server
Revert: ok
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665137 13f79535-47bb-0310-9956-ffa450edef68
Summary: If you want requests processed outside of the I/O thread
Reviewed By: jake luciani, aditya
Test Plan: nb-main.cpp, in the test folder
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665132 13f79535-47bb-0310-9956-ffa450edef68
Summary: I'm tired of getting output from thrift via perror AND exceptions, so
this class allows the client to set an alternate (or empty) handler for error
output
Reviewed By: mcslee
Test Plan: I ran on the worker with the default, got output via perror, then
overloaded with my own function and got output via syslog and then NULL
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665131 13f79535-47bb-0310-9956-ffa450edef68
Added explicit detached getter and setter
Modified PosixThreadFactory::~PThread:
Check for join failing and don't transition to detached_ state if it does. Potential thread-handle leak for
threads created joinable who aren't referenced by any external thread. Solution for now has to be
"DONT DO THAT", the clever approach doesn't always work.
Added ThreadFactoryTests.floodNThreads:
Loop M times for N threads where M x N is bigger than 32K to verify that detached threads can be created
ad infinitum.
Reviewed By: mcslee
Revert Plan: revertible
Test Plan: concurrency_test thread-factory passes
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665130 13f79535-47bb-0310-9956-ffa450edef68
Pay attention to detached flags. If thread is create non-detached and has not been joined when all references are given up,
(ie boost::share_ptr calls ~PThread) do the join in the destructor to prevent thread ids from being leaked.
Modified ThreadFactoryTests.reapNThreads:
Loop M times for M threads where M x N is bigger than 32K to verify that thread ids aren't leaked
Modified TimerManager.cpp:
Removed debug messages.
Reviewed By: mcslee
Revert Plan: revertible
Test Plan: concurrency_test thread-factory passes
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665129 13f79535-47bb-0310-9956-ffa450edef68
Summary: Replace with int64_t and don't worry about what architecture machine you're on, the typedefed int64_t will do the right thing.
Reviewed By: aditya, marc
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665123 13f79535-47bb-0310-9956-ffa450edef68
Summary:
- turns out that EAGAIN can be returned both when there is a transmission timeout and when there
is a lack of system resources.
This diff has a hacky fix for respecting a user specified read timeout.
Reviewed By: Steve Grimm, Marc, Slee
Test Plan:
- Tested by trying to crash an srp machine
Revert Plan: No.
Notes:
- Also added functionality to allow users to specify the max number of recv retries (in the case
when EAGAIN is returned due to a lack of system resources)
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665121 13f79535-47bb-0310-9956-ffa450edef68
Throw TimedOutException on wait timeout so caller can distinguish between timeout and event.
Modified facebook::thrift::concurrency::PthreadThread.start:
Throw SystemrResourceException on any pthread_* function call failure rather than asserting 0.
Added facebook::thrift::concurrency::Thread.id() and facebook::thrift::concurrency::ThreadFactory.currentThreadId():
Return thread-id of thread and current thread respectively. Needed for reentrancy tests in ThreadManager
Added facebook::thrift::concurrency::ThreadManager.pendingTaskCountMaxN
Modified facebook::thrift::concurrency::ThreadManager.add():
Now support a maximum pending task count and block if the current pending task count is max.
If timeout is specified for add, TimedOutException is thrown if pending task count doesn't decrease
in the timeout interval. If add() is called by a ThreadManager worker thread and the task cannot
be added, a TooManyPendingTasksException is thrown rather than blocking, since deadlocks can ensue
if worker threads block waiting for works threads to complete tasks.
Reviewed By: mcslee, aditya
Revert Plan: revertible
Test Plan: concurrency/test/ThreadManagerTests.h
run concurrency-test thread-manager
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665120 13f79535-47bb-0310-9956-ffa450edef68
Summary: SO_REUSEADDR doesn't always work, and some of my servers have
expensive init routines that I really want to run before it becomes available
to services, so I want thrift to retry the bind several times at varied
intervals before giving up
Reviewed By: slee
Test Plan: running a service in GDB with no problems binding, with another
service bound to the port for all the retries, another service bound to the
port for only some of the retries
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665109 13f79535-47bb-0310-9956-ffa450edef68
Summary: Identical to string in all languages except Java. Java String is NOT binary-safe, so we need to use raw byte[] in that case. PHP/RUBY/Python strings are all binary safe, and C++ std::string works fine and manages memory for you so it's the safest route. Java just needs this tweak.
Reviewed By: aditya
Test Plan: Use "binary" as a type instead of String.
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665099 13f79535-47bb-0310-9956-ffa450edef68
Summary:
- Make sure that buffers are always emptied when destructor is called (or when file is closed)
- make flush a no-op if file is not open
- dont allow new events to be enqueued if the file is in the process of being closed
Reviewed By: slee, jwang
Test Plan: tested with building search indices on top of thrift logs
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665094 13f79535-47bb-0310-9956-ffa450edef68
Summary: Not including config.h means we never get the HAVE_CLOCK_GETTIME #define, breaking the build with a duplicate function definition on machines that have a clock_gettime() function in the standard library.
Reviewed By: slee
Test Plan: it compiles, ship it
Revert: ok
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665087 13f79535-47bb-0310-9956-ffa450edef68
Summary:
- cannot assume that assert.h is defined
Reviewed By: marc k
Test Plan: everything compiles
Notes:
- need to reflect these changes in libfacebook/fbthread
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665073 13f79535-47bb-0310-9956-ffa450edef68
- StatsProcessor can be used to print events, or keep track of event frequency
- PeekProcessor is used to examine data in a thrift event, prior to passing it along to an underlying processor
- TPipedFileReaderTransport and its factory are used to pipe a TFileReaderTransport (which TFileProcessor requires)
Also fixed some bugs in TFileTransport - next flush time was overflowing and not always being reset
Reviewed by: aditya, mcslee
Test Plan: Tested using various thrift clients (scribe, falcon) and gdb in sandbox and on dev008.
git-svn-id: https://svn.apache.org/repos/asf/incubator/thrift/trunk@665066 13f79535-47bb-0310-9956-ffa450edef68
