A previous fix to make close() occur later can cause socket reuse by servers to fail as previous sockets are left asynchronously open.
This change:
- adds a callback to TCP server shutdown to signal that the server is completely shutdown
- wait for that callback before destroying listeners in the server (and before destroying the server)
- handles fallout
A previous fix to make close() occur later can cause socket reuse by servers to fail as previous sockets are left asynchronously open.
This change:
- adds a callback to TCP server shutdown to signal that the server is completely shutdown
- wait for that callback before destroying listeners in the server (and before destroying the server)
- handles fallout
-) You can't assume bash is installed. Scripts needs to be cleaned out of bashisms.
-) You can't assume python is in /usr/bin. Use env instead.
-) AF_INET is in sys/socket.h
-) Added port_platform's basic structure for FreeBSD, based off Darwin.
-) FreeBSD doesn't have and doesn't need libdl for OpenSSL.
We now pass down pointers to closures instead of (callback, arg) pair
elements separately. This allows us to store one word atomically, fixing
a race condition.
All call sites have been updated to the new API. No new allocations are
incurred. grpc_fd_state is deleted to avoid any temptation to ever add
anything there again.
Allow binding a different completion queue to each registered method.
This will allow multiplexing for the C++ server between sync & async
methods more easily.
This is a multipoller based on epoll rather than poll.
Note that this implementation is aimed at correctness rather than
performance, although it should immediately have better scalability to
large numbers of FDs, both due to epoll's O(1) sized API and due to not
needing to wake up polling threads to do interest set changes.
One notable difference here is that we directly attach a wakeup fd
rather than using the freelisting kick mechanism that the poll() based
implementations use, because modifying the epoll set to use a different
kick fd each time isn't free.
This introduces the wakeup fd interface, corresponding approximately to
the existing Google version, complete with a ported giant detailed usage
comment.
The implementation has two layers, "specialized" and "fallback". The
specialized layer is intended to be a suitable platform specific
implementation like eventfd, whereas "fallback" is probably pipe, with
runtime detection of whether the specialized version works on this
system (currently stubbed out).
This change pulls out a separate pollset_kick module, which currently
uses a freelist of pipes dynamically assigned to pollsets when they
enter polling rather than the previous racy sharding mechanism.
We ultimately may wish to eliminate the dynamic assignment for multipoll
sets, but this should be sufficient for the moment.
Fixed any exposed bugs across the stack.
Add a poll() based implementation. Heavily leverages pollset infrastructure to allow small polls to be the norm. Exposes a mechanism to plug in epoll/kqueue for platforms where we have them.
Simplify iomgr callbacks to return one bit of success or failure (instead of the multi valued result that was mostly unused previously). This will ease the burden on new implementations, and the previous system provided no real value anyway.
Removed timeouts on endpoint read/write routines. This simplifies porting burden by providing a more orthogonal interface, and the functionality can always be replicated when desired by using an alarm combined with endpoint_shutdown. I'm fairly certain we ended up with this interface because it was convenient to do from libevent.
Things that need attention still:
- adding an fd to a pollset is O(n^2) - but this is probably ok given that we'll not
use this for multipolling once platform specific implementations are added.
- we rely on the backup poller too often - especially for SSL handshakes and for client
connection establishment we should have a better mechanism ([] []
- Linux needs to use epoll for multiple fds, FreeBSD variants (including
Darwin) need to use kqueue. ([] []
- Linux needs to use eventfd for poll kicking. ([]
Change on 2015/01/07 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=83461069
Fixed any exposed bugs across the stack.
Add a poll() based implementation. Heavily leverages pollset infrastructure to allow small polls to be the norm. Exposes a mechanism to plug in epoll/kqueue for platforms where we have them.
Simplify iomgr callbacks to return one bit of success or failure (instead of the multi valued result that was mostly unused previously). This will ease the burden on new implementations, and the previous system provided no real value anyway.
Removed timeouts on endpoint read/write routines. This simplifies porting burden by providing a more orthogonal interface, and the functionality can always be replicated when desired by using an alarm combined with endpoint_shutdown. I'm fairly certain we ended up with this interface because it was convenient to do from libevent.
Things that need attention still:
- adding an fd to a pollset is O(n^2) - but this is probably ok given that we'll not
use this for multipolling once platform specific implementations are added.
- we rely on the backup poller too often - especially for SSL handshakes and for client
connection establishment we should have a better mechanism ([] []
- Linux needs to use epoll for multiple fds, FreeBSD variants (including
Darwin) need to use kqueue. ([] []
- Linux needs to use eventfd for poll kicking. ([]
Change on 2015/01/07 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=83461069
This change implements a platform independent alarm manager in alarm.c.
It's integrated with iomgr, and some tests are cleaned up.
The alarm implementation itself is a fairly direct port of LazyAlarmList from eventmanager.
SpinLock has been replaced for now with gpr_mu, and other atomic operations have been dropped (again, for now).
A majority of tests have been ported.
Change on 2014/12/19 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=82551363
Fold endpoint interface into iomgr, move secure_endpoint into security/.
This will make it easier for endpoint to rely on some iomgr defined types (like
pollset).
Change on 2014/12/17 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=82338036
This will allow combining of gpr_now() calls, and make it easier to test the
alarm code in the future.
Change on 2014/12/16 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=82246768
Clients stay connected to a server after it shutdowns until all active calls
have completed, and then they drop.
After a GOAWAY or a disconnect, clients will attempt to re-resolve and
reconnect to the server.
Change on 2014/12/15 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=82178515
This is a trivial C89-ification of the []2 implementation of the same idea.
Indeed the implementation files and tests have been branched from the [] versions.
Change on 2014/12/15 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=82168822
This change removes a corner case that needs to be handled (what happens if you
add an alarm twice?). This corner case was never used, nor tested.
Change on 2014/12/15 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=82168179
Move eventmanager and platform dependent endpoint functionality into a single
library called 'iomgr'.
This is primarily to prepare for a Windows port - where posix socket semantics
lead to poor quality code.
Mostly this is a code movement CL, with some small changes to help prepare the
way for porting:
- em style fd objects can only be held internally in iomgr, and own their memory
- added grpc_iomgr_create_endpoint_pair() to accomodate the common pattern of
creating a tcp endpoint from the output of socketpair - this will help keep
our tests portable
- separated em alarm interface into a separate file, as this part of event
manager is needed higher up the stack
- made the eventmanager bits a true singleton, simplifying API's across the
stack as there's no longer a reason to carry a pointer there.
Initial design document is here:
https://docs.google.com/document/d/1VmafcHvvrP5kwtQkz84R5yXF7u7fW-9Pn0bkSUQHDt8/edit?disco=AAAAARNByxg
Change on 2014/12/09 by ctiller <ctiller@google.com>
-------------
Created by MOE: http://code.google.com/p/moe-java
MOE_MIGRATED_REVID=81716456
Craig Tiller
Nicolas Noble
Craig Tiller
Nicolas Noble
David Klempner
Yang Gao