Feature #2989
closedMinimal UDP permanent face
Added by Junxiao Shi over 9 years ago. Updated about 9 years ago.
Description
Implement a minimal UDP permanent face.
In this minimal implementation,
UdpFace
constructor acceptsFacePersistency persistency
parameter in place ofbool isOnDemand
.- If
persistency == FacePersistency::PERMANENT
, the face ignores all socket errors after successful creation.
Errors during the initial face creation should be reported as usual. - There is no UP/DOWN state transitions.
An operator who decides to use a permanent face must take care to choose a strategy that is able to try multiple faces (ie. notbest-route
).
This Feature also includes the following changes:
Face::isOnDemand
andFace::setOnDemand
are replaced withFace::getPersistency
andFace::setPersistency
(also,bool m_isOnDemand
is replaced withFacePersistency m_persistency
); all usages of those methods shall be replacedProtocolFactory::createFace
needs aFacePersistency persistency
parameter.
If aProtocolFactory
subclass does not support the specifiedpersistency
, throwProtocolFactory::Error
.UdpFactory::createFace
should supportFacePersistency::PERSISTENT
andFacePersistency::PERMANENT
.FaceManager::createFace
passesFacePersistency::PERSISTENT
toProtocolFactory::createFace
.
To allow an operator to use UDP permanent faces, management and nfdc
changes are needed, but they are not part of this issue.
Updated by Junxiao Shi over 9 years ago
- Related to Task #2491: Design permanent faces added
Updated by Junxiao Shi over 9 years ago
- Blocks Feature #2991: FaceManager: specify face persistency added
Updated by Davide Pesavento over 9 years ago
Junxiao Shi wrote:
- If
persistency == FacePersistency::PERMANENT
, the face ignores socket errors.
Which errors are to be ignored and which ones are not? For example failures to create a socket should not be ignored in my opinion, as they are usually a symptom of a bigger problem.
- The face transmits packets whenever the underlying socket allows immediate transmission.
I have no idea what this sentence means.
Updated by Junxiao Shi over 9 years ago
- Description updated (diff)
All socket errors must be ignored, because the goal is to allow the face operate unattended.
However, errors are allowed during creation, as long as those errors are reported before FaceCreatedCallback
is invoked.
For now, this means if socket
or connect
fails, the face will be closed.
After #1285, the face can also be closed if tunnel authentication fails.
Updated by Junxiao Shi over 9 years ago
- Assignee set to Yukai Tu
20150629 conference call approves this issue, and decides to assign it to @Yukai.
Updated by Yukai Tu over 9 years ago
Junxiao Shi wrote:
ProtocolFactory::createFace
needs aFacePersistency persistency
parameter.
If aProtocolFactory
subclass does not support the specifiedpersistency
, throwProtocolFactory::Error
.
Should I rewrite all subclass to support the persistency?
Updated by Junxiao Shi over 9 years ago
Reply to note-7:
In all ProtocolFactory
subclasses except UdpFactory
, the createFace
method shall be rewritten to accept persistency
parameter. If the argument value is not PERSISTENT
, the implementation throws Error
type of that subclass.
Updated by Yukai Tu over 9 years ago
UdpFactory don't need to accept persistency parameter?
Updated by Junxiao Shi over 9 years ago
Answer to note-9:
UdpFactory::createFace
certainly needs to accept persistency
parameter, but it doesn't throw exception for PERMANENT
. Instead:
UdpFactory::createFace should support FacePersistency::PERSISTENT and FacePersistency::PERMANENT.
Updated by Davide Pesavento over 9 years ago
Junxiao Shi wrote:
If the argument value is not
PERSISTENT
, the implementation throwsError
type of that subclass.
if the value is not PERSISTENT
? I guess you meant the opposite...
Also, I suggest to implement this feature with at least two separate changesets. First, refactor/change the existing code to use FacePersistency
instead of bool
. Then implement the actual permanent UDP face.
Updated by Junxiao Shi over 9 years ago
Answer to note-11:
I confirm note-8 is accurate, not the opposite.
Using multiple commits for the same issue is acceptable.
Updated by Davide Pesavento over 9 years ago
Junxiao Shi wrote:
I confirm note-8 is accurate, not the opposite.
What about on-demand faces then? (actually I didn't really mean the opposite in note 11... my point was that we should throw an error only if face persistency is permanent, except for udp unicast)
Updated by Junxiao Shi over 9 years ago
Answer to note-13:
ProtocolFactory::createFace
is used to initiate an outgoing connection; FacePersistency::ON_DEMAND
is for incoming connection only.
Previously, ProtocolFactory::createFace
does not have a parameter to indicate whether the face is on-demand or persistent, and all faces are created as persistent.
This issue adds permanent face to UdpFactory::createFace
.
It does not change that fact that on-demand is not supported by ProtocolFactory::createFace
.
Updated by Yukai Tu over 9 years ago
Just to make sure that I got it:
- Rewrite ProtocolFactory::createFace to support a FacePersistency persistency parameter.
- All ProtocolFactory subclasses should be rewrite. But only UdpFactory::createFace support FacePersistency::PERSISTENT and FacePersistency::PERMANENT.
- Original onDemand system should be discarded.
- Other subclasses only accept FacePersistency::PERSISTENT, passing FacePersistency::PERMANENT will throw Error.
Updated by Junxiao Shi over 9 years ago
I neither confirm nor deny note-16. Please refer to the issue description.
Updated by Junxiao Shi over 9 years ago
- Assignee changed from Yukai Tu to Junxiao Shi
- % Done changed from 0 to 40
http://gerrit.named-data.net/2185
I notice Face::setOnDemand
is problematic and needs a change.
I'll update the design for this.
Updated by Junxiao Shi over 9 years ago
- Description updated (diff)
- Assignee changed from Junxiao Shi to Yukai Tu
- Estimated time changed from 3.00 h to 5.00 h
As recommended in note-11, the Feature should be completed in 3 commits:
face: replace isOnDemand with FacePersistency
* replace `Face::isOnDemand` and `Face::setOnDemand` with API based on `Face::getPersistency` and `Face::setPersistency`
* make necessary changes in subclasses and callers
face: ProtocolFactory::createFace with FacePersistency
* in `ProtocolFactory::createFace`, add `persistency` parameter
* make necessary changes in subclasses and callers
* ALL subclasses, including `UdpFactory`, shall throw exception when `persistency != PERSISTENT` at this commit
face: minimal UDP permanent face
* change `UdpFace` constructor to accept `persistency`
* change `UdpFactory::createFace` to allow `PERMANENT`
* implement minimal UDP permanent face semantics in `UdpFace`
Updated by Yukai Tu over 9 years ago
Junxiao Shi wrote:
* ALL subclasses, including `UdpFactory`, shall throw exception when `persistency != PERSISTENT` at this commit * change `UdpFactory::createFace` to allow `PERMANENT`
If throw exception when persistency != PERSISTENT
, how to allow PERMANENT
?
Updated by Junxiao Shi over 9 years ago
Answer to note-21:
There is no conflict in note-20 design. The first quoted statement applies to the second commit, while the second quoted statement applies to the third commit which adds a new feature onto the second commit.
Updated by Yukai Tu over 9 years ago
After cloning project from gerrit, I run ./waf configure --with-tests then ./waf, and it always builds fail. Is there any wrong with the codes?
../tests/core/segment-publisher.t.cpp: 在成员函数‘virtual std::size_t nfd::tests::SegmentPublisherTester<N>::generate(ndn::encoding::EncodingBuffer&)’中:
../tests/core/segment-publisher.t.cpp:76:81: 错误: ‘prependNonNegativeIntegerBlock’的实参不依赖模板参数,所以‘prependNonNegativeIntegerBlock’的声明必须可用 [-fpermissive]
../tests/core/segment-publisher.t.cpp:76:81: 附注: (如果您使用‘-fpermissive’,G++ 会接受您的代码,但是允许使用未定义的名称是不建议使用的风格)
../tests/core/segment-publisher.t.cpp: 在成员函数‘std::size_t nfd::tests::SegmentPublisherTester<N>::generate(ndn::encoding::EncodingBuffer&) [with long int N = 10000l, std::size_t = long unsigned int, ndn::encoding::EncodingBuffer = ndn::encoding::EncodingImpl<true>]’中:
../tests/core/segment-publisher.t.cpp:171:1:从此处实例化
../tests/core/segment-publisher.t.cpp:76:9: 错误: ‘prependNonNegativeIntegerBlock’在此作用域中尚未声明
../tests/core/segment-publisher.t.cpp:76:9: 附注: 建议的替代:
/usr/local/include/ndn-cxx/encoding/block-helpers.hpp:55:1: 附注: ‘ndn::prependNonNegativeIntegerBlock’
../tests/core/segment-publisher.t.cpp: 在成员函数‘std::size_t nfd::tests::SegmentPublisherTester<N>::generate(ndn::encoding::EncodingBuffer&) [with long int N = 100l, std::size_t = long unsigned int, ndn::encoding::EncodingBuffer = ndn::encoding::EncodingImpl<true>]’中:
../tests/core/segment-publisher.t.cpp:171:1:从此处实例化
../tests/core/segment-publisher.t.cpp:76:9: 错误: ‘prependNonNegativeIntegerBlock’在此作用域中尚未声明
../tests/core/segment-publisher.t.cpp:76:9: 附注: 建议的替代:
/usr/local/include/ndn-cxx/encoding/block-helpers.hpp:55:1: 附注: ‘ndn::prependNonNegativeIntegerBlock’
../tests/core/segment-publisher.t.cpp: 在成员函数‘std::size_t nfd::tests::SegmentPublisherTester<N>::generate(ndn::encoding::EncodingBuffer&) [with long int N = 10l, std::size_t = long unsigned int, ndn::encoding::EncodingBuffer = ndn::encoding::EncodingImpl<true>]’中:
../tests/core/segment-publisher.t.cpp:171:1:从此处实例化
../tests/core/segment-publisher.t.cpp:76:9: 错误: ‘prependNonNegativeIntegerBlock’在此作用域中尚未声明
../tests/core/segment-publisher.t.cpp:76:9: 附注: 建议的替代:
/usr/local/include/ndn-cxx/encoding/block-helpers.hpp:55:1: 附注: ‘ndn::prependNonNegativeIntegerBlock’
../tests/core/segment-publisher.t.cpp: 在成员函数‘std::size_t nfd::tests::SegmentPublisherTester<N>::generate(ndn::encoding::EncodingBuffer&) [with long int N = 0l, std::size_t = long unsigned int, ndn::encoding::EncodingBuffer = ndn::encoding::EncodingImpl<true>]’中:
../tests/core/segment-publisher.t.cpp:171:1:从此处实例化
../tests/core/segment-publisher.t.cpp:76:9: 错误: ‘prependNonNegativeIntegerBlock’在此作用域中尚未声明
../tests/core/segment-publisher.t.cpp:76:9: 附注: 建议的替代:
/usr/local/include/ndn-cxx/encoding/block-helpers.hpp:55:1: 附注: ‘ndn::prependNonNegativeIntegerBlock’
Updated by Junxiao Shi over 9 years ago
Reply to note-24:
It seems that you have conflicting versions of ndn-cxx and NFD.
Please ensure both are latest version and try again.
Updated by Yukai Tu over 9 years ago
What should I do to rebase? I used command "git rebase" but it doesn't work.
Updated by Davide Pesavento over 9 years ago
Are you working on a separate branch? Did you pull from the upstream master branch before rebasing?
Updated by Yukai Tu over 9 years ago
I working on the master branch. And git rebase command seems don't match the requirements. What command should I use?
Updated by Davide Pesavento over 9 years ago
Working on the master branch is a terrible idea as it doesn't let you work on multiple patchsets at the same time.
I believe you can use git pull --rebase
on the master branch to fix the current situation (if you haven't already done a merge with upstream), but I strongly suggest using separate local branches next time.
Updated by Junxiao Shi over 9 years ago
- Related to Task #2696: Document the behavior of persistent and on-demand faces added
Updated by Yukai Tu over 9 years ago
How about adding ignore() into BaseFixture?
Updated by Yukai Tu over 9 years ago
If the socket is closed or It receives the boost::asio::error::eof, should I reconnect to the remote endpoint? If it still fails, what should permanent face do?
Updated by Davide Pesavento over 9 years ago
Yukai Tu wrote:
If the socket is closed or It receives the boost::asio::error::eof, should I reconnect to the remote endpoint?
If the socket is closed, yes, it needs to be reopened and reconnected. If error::eof
results in the socket being closed, then it should be reopened and reconnected as well, otherwise just ignore the eof and continue. This applies to all errors that do not leave the socket in a bad/inconsistent/unusable state.
If it still fails, what should permanent face do?
If what fails? The connection attempt?
Note that since UDP is connection-less, a connect
operation does not in fact perform any network operations. Therefore I don't think that a connect
on a UDP socket can actually fail for network problems. It can fail for "local" reasons (e.g. resource exhaustion), but in that case I believe retrying would be pointless and you should let the face fail. The issue description is not clear on this point however, so let's wait for Junxiao's comment.
Updated by Junxiao Shi over 9 years ago
It can fail for "local" reasons (e.g. resource exhaustion), but in that case I believe retrying would be pointless and you should let the face fail.
No, a permanent face can never fail. This is a strong guarantee.
It could be DOWN.
In the minimal implementation, there's no explicit DOWN state.
The face can be kept in an unusable state (not able to send or receive).
Before each attempt to send, retry connect.
Updated by Davide Pesavento over 9 years ago
Ok then. What if bind
fails though? Or async_receive
?
Btw, maybe we should use sendto/recvfrom
instead of connect
+ send/receive
?
Updated by Yukai Tu over 9 years ago
Davide Pesavento wrote:
Ok then. What if
bind
fails though? Orasync_receive
?
If bind fails, we can ignore the error and reconnect until next attempt to send.
Btw, maybe we should use
sendto/recvfrom
instead ofconnect
+send/receive
?
Maybe 'connect' + 'send/receive' is acceptable.
Updated by Davide Pesavento over 9 years ago
Yukai Tu wrote:
Davide Pesavento wrote:
Ok then. What if
bind
fails though? Orasync_receive
?If bind fails, we can ignore the error and reconnect until next attempt to send.
"reconnect" has nothing to do with bind
. I'm talking about the receive path.
Btw, maybe we should use
sendto/recvfrom
instead ofconnect
+send/receive
?Maybe 'connect' + 'send/receive' is acceptable.
Of course it's acceptable. The point is: which way is simpler? (connect
, being an additional call, is potentially another source of errors).
@Alex/Junxiao: what were the original reasons to prefer connect
+ send/receive
?
Updated by Alex Afanasyev over 9 years ago
@Alex/Junxiao: what were the original reasons to prefer
connect
+send/receive
?
To avoid reimplementing dst/port-based dispatch manually.
Updated by Davide Pesavento over 9 years ago
Alex Afanasyev wrote:
@Alex/Junxiao: what were the original reasons to prefer
connect
+send/receive
?To avoid reimplementing dst/port-based dispatch manually.
ok, makes sense. It's probably easier to keep the connect
then.
Updated by Junxiao Shi over 9 years ago
All code changes are merged. Assignee also needs to update NFD Developer Guide before this issue can close.
Updated by Yukai Tu about 9 years ago
How can I update NFD Developer Guide? Thanks.
Updated by Junxiao Shi about 9 years ago
- Description updated (diff)
- Status changed from In Progress to Closed
- % Done changed from 40 to 100
Documentation work is moved to #3173 because face refactoring is coming soon.