1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 <2025> | Index | 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 <2025> |
<== Date ==> | <== Thread ==> |
---|
Subject: | Re: Help with bumpless IOC reboot, record linking and initialization concepts |
From: | Mark Rivers via Tech-talk <tech-talk at aps.anl.gov> |
To: | Michael Davidsaver <mdavidsaver at gmail.com>, "Dmitry Yu. Bolkhovityanov" <D.Yu.Bolkhovityanov at inp.nsk.su> |
Cc: | EPICS Tech Talk <tech-talk at aps.anl.gov> |
Date: | Sun, 21 Sep 2025 16:49:21 +0000 |
Yes.
asynManager provides the following methods that are used use for connection management:
asynStatus (*exceptionConnect)(asynUser *pasynUser);
asynStatus (*exceptionDisconnect)(asynUser *pasynUser);
asynStatus (*autoConnect)(asynUser *pasynUser,int yesNo);
asynStatus (*isConnected)(asynUser *pasynUser,int *yesNo);
asynStatus (*isAutoConnect)(asynUser *pasynUser,int *yesNo);
Those are documented here:
The asynCommon interface provides the following methods that asynManager and asyn clients use for connection management:
typedef struct asynCommon {
...
asynStatus (*connect)(void *drvPvt,asynUser *pasynUser);
asynStatus (*disconnect)(void *drvPvt,asynUser *pasynUser);
}asynCommon;
These methods must be implemented by all drivers. They are documented here:
It is always good practice to load an asyn record for each port. That provides the following PVs:
Those are documented here:
The asyn TCP driver (drvAsynIPPort) implements connection management. It connects (opens the socket) when the pasynCommon->connect() method is called. If the socket open is successful, then the driver calls pasynManager->exceptionConnect to indicate that
it is connected. If isAutoConnect is true then asynManager will call this method when the port is created, and whenever it is disconnected and a client queues an I/O request. If isAutoConnect is false then pasynCommon->connect() method must be called by
the client. Whenever the driver detects that the socket is disconnected it calls pasynManager->exceptionDisconnect to indicate that state.
It can take a long time for the OS to close a socket for a non-responsive device, and I/O operations only return the disconnected status once that happens. The asynOption disconnectOnReadTimeout can be used to force the driver to disconnect whenever an I/O
operation times out. Michael Davidsaver has asked whether that should be the default behavior. I would be interested to hear whether that would be acceptable, or would break expected behavior.
Mark
From: Tech-talk <tech-talk-bounces at aps.anl.gov> on behalf of Dmitry Yu. Bolkhovityanov via Tech-talk <tech-talk at aps.anl.gov>
Sent: Sunday, September 21, 2025 2:39 AM To: Michael Davidsaver <mdavidsaver at gmail.com> Cc: EPICS Tech Talk <tech-talk at aps.anl.gov> Subject: Re: Help with bumpless IOC reboot, record linking and initialization concepts Hi Michael!
On Sat, 20 Sep 2025, Michael Davidsaver via Tech-talk wrote: > On 9/19/25 00:35, Marco Filho wrote: >> Ah, I think I didn't properly understand this email on the first read. >> >> > Generally, I have found that the more effective way to coordinate >> > connect/disconnect is at the driver level. Generally with a special >> > device support to drive on-connect processing. IMHO, it isn't a "more effective" but is rather the ONLY reliable way. As only the driver code, which issues a connect() call and is directly notified about connect success or failure, thus has first-hand knowledge of the connection status and is able to perform connect/disconnect. All the other ways use some form of side effects, which are unreliable and can change with time. >> You mean disconnecting from the actual equipment before the PVs are >> initialized and then only connecting once all the links are properly >> initialized? >> If so, how do you detect at driver level that that has happened? > > I think you have right idea, although the other way around. My thinking is > that a driver initializes into a latched "error" state, which is cleared by > processing a record. I like this approach because it naturally leads to > handling the initial connection the same as a re-connect. For me it seems obvious to use 2 boolean PVs: 1. "isConnected", which is SCAN="I/O intr" and is updated by the driver upon change of connection state to be 1 (connected) or 0 (disconnected); initially value is 0. (It can be naturally monitored by code and DB links; multiple "isConnected" values can be ANDed). 2. "doConnect", which can be written to, =1 means "connect if not connected" and =0 means "disconnect". Isn't such a thing already done somewhere in asyn, streamDevice etc.? |