pcp
[Top] [All Lists]

Re: [pcp] pcp updates: PM_ERR_TIMEOUT protocol changes

To: Ken McDonell <kenj@xxxxxxxxxxx>
Subject: Re: [pcp] pcp updates: PM_ERR_TIMEOUT protocol changes
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Mon, 1 Aug 2016 20:41:28 -0400 (EDT)
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <20160801101257.3C8855A00D5@kenj>
References: <20160801101257.3C8855A00D5@kenj>
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Thread-index: AdJZCiL1PAVJmJ47PbN7TNfgPQeWCA==
Thread-topic: pcp updates: PM_ERR_TIMEOUT protocol changes
Hi Ken,

----- Original Message -----
> [...]
> Given the severity of the issue, it is hard to imagine these changes
> making the situation worse ... and they demonstrably make it better
> in a range of QA tests built to simulate the BZ 1158 failure signature.
> 
> There are still some __getPDU() uses to be analyzed, but these all
> relate to context creation and should a similar problem occur there
> and be undetected the changes here will catch it almost immediately
> the real client-pmcd traffic starts.

I guess the other place to still consider is the pmlc PDUs, but they
are never multiplexed so probably not an area to really worry about.

I can't see any issues here.  One slight tweak that might be worth
doing (maybe later - its an internal routine only) would be to have
the new __pmCloseChannel routine pull in the two lines that always
follow it, i.e. fixing up the error code to be one of TIMEOUT/IPC,
and returning that, instead of open-coding that logic for each PDU.

cheers.

--
Nathan

<Prev in Thread] Current Thread [Next in Thread>