Hehe, thanks for introduction. :-) I admit, I joined the discussion to help to make replicated journal more robust. This feature is doing that already and thanks for implementing it. I'm happy that more people can share their ideas and different approaches.
You're right, adding backup as default would stop live server if no other IP address would be provided to the check list and backup would be killed/crashed. It's not possible to distinguish between network failure and crash of remote participant. An IF check could be provided that list contains at least one other IP address but this is getting over-engineered which is against the original idea to have this simple. > I will add an option to add backups to network health check, but I > won't do it by default. Ok, let's not do assumptions here. Thanks, Mirek ----- Original Message ----- > From: "Clebert Suconic" <[email protected]> > To: [email protected] > Sent: Monday, November 28, 2016 3:34:29 PM > Subject: Re: [DISCUSS] ARTEMIS-863 - Network Health Check > > First, before I continue.. let me introduce Miroslav to everybody: > > Miroslav is a QA engineer at Red Hat for EAP, Messaging and so on, and > to be fair.. he is one of the best I have ever worked with. He is > really good on finding issues.. I have an internal joke that I'm > always afraid when I see a JIRA opened by Mirek. so I really respect > his opinion. > > It is great that we are having an open discussion here. and I welcome > you to always be part of open discussions here at the dev list. > > > So, back to the issue: > > I will add an option to add backups to network health check, but I > won't do it by default. You won't always have ICMP open. and I won't > add anything without a third address added. Because losing the backup > would lose the whole live as well (in case of a power outage). > > The intent of the PING is to avoid network failures on the box. E.G.. > If you pull the cable of the box. > > If you make weird routes to your system, then you are on purpose > splitting your network in such way you don't have how to fix. You need > redundant routes at the system... just like any other networking. (not > to cut them) > > The user should know his topology and should always provide IPs that > will be helpful on validating the network. The user also need to > consider his network topology on the solution. > > On Fri, Nov 25, 2016 at 6:20 AM, Miroslav Novak <[email protected]> wrote: > > Yes, this could be configurable. I would vote to have this enabled by > > default as in case with replicated journal. This will prevent situation > > where live and backup are active at the same time. I don't see use for > > this feature in simple cluster. > > > > It seems that check list cannot contain IP to itself as such address will > > be always reachable. > > > > Mirek > > > > ----- Original Message ----- > >> From: "Clebert Suconic" <[email protected]> > >> To: [email protected] > >> Sent: Thursday, November 24, 2016 2:53:30 PM > >> Subject: Re: [DISCUSS] ARTEMIS-863 - Network Health Check > >> > >> We can add a few properties to such to control add backup or add other > >> members. > >> > >> I will think about something next week after the holiday. > >> > >> On Thu, Nov 24, 2016 at 3:18 AM Miroslav Novak <[email protected]> wrote: > >> > >> > For this to work properly, the check list of IPs/hostnames must contain > >> > also live/backup pair. I mean that live should have in its checklist IP > >> > of > >> > backup and backup of its live server. I think there is problem for the > >> > user > >> > to manually add this to the "check list", especially if it's using > >> > discovery - JGroups UDP or Netty UDP discovery - for finding servers. > >> > Can > >> > IP/hostnames be added to the checklist automatically? Otherwise this > >> > will > >> > be very static. wdyt? > >> > > >> > Mirek > >> > > >> > ----- Original Message ----- > >> > > From: "Clebert Suconic" <[email protected]> > >> > > To: [email protected] > >> > > Sent: Tuesday, November 22, 2016 5:36:44 PM > >> > > Subject: Re: [DISCUSS] ARTEMIS-863 - Network Health Check > >> > > > >> > > I changed the implementation to accept IPV6. > >> > > > >> > > Unfortunately, an IPV6 Address is not returning true through > >> > > isReachable. Perhaps I really need root access fot the ICMP to work on > >> > > IPV6 address. > >> > > > >> > > > >> > > Anyway, I am testing for IPV6 and using ping6 now on my local copy.. I > >> > > will update the branch before the end of the day, and I should send a > >> > > PR. > >> > > > >> > > On Tue, Nov 22, 2016 at 11:25 AM, Matt Pavlovich <[email protected]> > >> > wrote: > >> > > > On 11/22/16 9:14 AM, Clebert Suconic wrote: > >> > > >> > >> > > >> On Tue, Nov 22, 2016 at 10:04 AM, Matt Pavlovich > >> > > >> <[email protected]> > >> > > >> wrote: > >> > > >>> > >> > > >>> Clebert- > >> > > >>> > >> > > >>> Can you clarify what you mean by: > >> > > >>> > >> > > >>> "..the server will shutdown.." > >> > > >> > >> > > >> > >> > > >> > >> > > >> the Network Checker will call server.stop(); > >> > > >> > >> > > >> > >> > > >> The process will stay alive, waiting the network to come back, when > >> > > >> is > >> > > >> back.. it will call server.start() again. > >> > > > > >> > > > Good deal. Sounds good! > >> > > > >> > > > >> > > > >> > > -- > >> > > Clebert Suconic > >> > > > >> > > >> > > > > -- > Clebert Suconic >
