Re: More entropy on bundle startup.
Thanks Guillaume! This will definitely help us as we find those holding onto stale service references. -Nick From: Guillaume Nodet <gno...@apache.org> Reply-To: "user@karaf.apache.org" <user@karaf.apache.org> Date: Thursday, July 14, 2016 at 3:11 PM To: user <user@karaf.apache.org> Subject: Re: More entropy on bundle startup. Try the load-test command inside the karaf console. It randomly starts / stop / refresh bundles with multiple threads in loop, so such things are quickly seen usually. 2016-07-13 22:56 GMT+02:00 Benson Margulies <ben...@basistech.com<mailto:ben...@basistech.com>>: Folks, We've had a couple of incidents of latent problems stemming from invalid assumptions on bundle start order. Everything seems to be fine, then some trivial change reveals that we've failed to ensure that service 'a' is available before component 'b' needs it. by and large, we use DS to get this right, but there are a few cases where it does not serve. I am wondering: is there some way to get _more_ randomness out of the startup process, to shake out mistakes like this? thanks, benson -- Guillaume Nodet Red Hat, Open Source Integration Email: gno...@redhat.com<mailto:gno...@redhat.com> Web: http://fusesource.com<http://fusesource.com/> Blog: http://gnodet.blogspot.com/
Re: More entropy on bundle startup.
Try the load-test command inside the karaf console. It randomly starts / stop / refresh bundles with multiple threads in loop, so such things are quickly seen usually. 2016-07-13 22:56 GMT+02:00 Benson Margulies: > Folks, > > We've had a couple of incidents of latent problems stemming from > invalid assumptions on bundle start order. Everything seems to be > fine, then some trivial change reveals that we've failed to ensure > that service 'a' is available before component 'b' needs it. by and > large, we use DS to get this right, but there are a few cases where it > does not serve. > > I am wondering: is there some way to get _more_ randomness out of the > startup process, to shake out mistakes like this? > > thanks, > benson > -- Guillaume Nodet Red Hat, Open Source Integration Email: gno...@redhat.com Web: http://fusesource.com Blog: http://gnodet.blogspot.com/
Re: More entropy on bundle startup.
We use DS for most of our services, but we have a few down at the activator level. I appreciate that the fixes to the bugs involve service tracking. The problem is _finding_ the bugs. Yes we use featuresBoot. featuresBootAsynchronous=false On Thu, Jul 14, 2016 at 9:56 AM, James Carmanwrote: > You are using featuresBoot? Do you have featuresBootAsynchronous=false in > your org.apache.karaf.features.cfg file? > > > On Thu, Jul 14, 2016 at 9:33 AM Benson Margulies > wrote: >> >> I don't think so, no. I do not do any dynamic installation. I use the >> Maven plugin to make an assembly with all the features I need. I then >> observe that the startup order is not deterministic from machine to >> machine, and it is particularly prone to change when I stop and start >> the container without clearing out the data directory. >> >> >> On Thu, Jul 14, 2016 at 1:39 AM, Jean-Baptiste Onofré >> wrote: >> > Hi Benson, >> > >> > I guess you are using the deploy folder, so the fileinstall deployer, >> > right >> > ? >> > >> > Regards >> > JB >> > >> > >> > On 07/13/2016 10:56 PM, Benson Margulies wrote: >> >> >> >> Folks, >> >> >> >> We've had a couple of incidents of latent problems stemming from >> >> invalid assumptions on bundle start order. Everything seems to be >> >> fine, then some trivial change reveals that we've failed to ensure >> >> that service 'a' is available before component 'b' needs it. by and >> >> large, we use DS to get this right, but there are a few cases where it >> >> does not serve. >> >> >> >> I am wondering: is there some way to get _more_ randomness out of the >> >> startup process, to shake out mistakes like this? >> >> >> >> thanks, >> >> benson >> >> >> > >> > -- >> > Jean-Baptiste Onofré >> > jbono...@apache.org >> > http://blog.nanthrax.net >> > Talend - http://www.talend.com
Re: More entropy on bundle startup.
You are using featuresBoot? Do you have featuresBootAsynchronous=false in your org.apache.karaf.features.cfg file? On Thu, Jul 14, 2016 at 9:33 AM Benson Margulieswrote: > I don't think so, no. I do not do any dynamic installation. I use the > Maven plugin to make an assembly with all the features I need. I then > observe that the startup order is not deterministic from machine to > machine, and it is particularly prone to change when I stop and start > the container without clearing out the data directory. > > > On Thu, Jul 14, 2016 at 1:39 AM, Jean-Baptiste Onofré > wrote: > > Hi Benson, > > > > I guess you are using the deploy folder, so the fileinstall deployer, > right > > ? > > > > Regards > > JB > > > > > > On 07/13/2016 10:56 PM, Benson Margulies wrote: > >> > >> Folks, > >> > >> We've had a couple of incidents of latent problems stemming from > >> invalid assumptions on bundle start order. Everything seems to be > >> fine, then some trivial change reveals that we've failed to ensure > >> that service 'a' is available before component 'b' needs it. by and > >> large, we use DS to get this right, but there are a few cases where it > >> does not serve. > >> > >> I am wondering: is there some way to get _more_ randomness out of the > >> startup process, to shake out mistakes like this? > >> > >> thanks, > >> benson > >> > > > > -- > > Jean-Baptiste Onofré > > jbono...@apache.org > > http://blog.nanthrax.net > > Talend - http://www.talend.com >
Re: More entropy on bundle startup.
What kind of services do you create? Blueprint, Component, direct Activator? I use Component, and the Component waits for the required components to be available and then starts. I have had issues in the past with badly coded direct Activators not starting in the correct order. I then found that I had to poll periodically in the Activator to see if the service was available now. Components took care of all that. Now this process means you have to use scr:list to determine that the Components did indeed start correctly, but we have monitors in place that verifies this for us now. On 07/14/2016 08:33 AM, Benson Margulies wrote: I don't think so, no. I do not do any dynamic installation. I use the Maven plugin to make an assembly with all the features I need. I then observe that the startup order is not deterministic from machine to machine, and it is particularly prone to change when I stop and start the container without clearing out the data directory. On Thu, Jul 14, 2016 at 1:39 AM, Jean-Baptiste Onofréwrote: Hi Benson, I guess you are using the deploy folder, so the fileinstall deployer, right ? Regards JB On 07/13/2016 10:56 PM, Benson Margulies wrote: Folks, We've had a couple of incidents of latent problems stemming from invalid assumptions on bundle start order. Everything seems to be fine, then some trivial change reveals that we've failed to ensure that service 'a' is available before component 'b' needs it. by and large, we use DS to get this right, but there are a few cases where it does not serve. I am wondering: is there some way to get _more_ randomness out of the startup process, to shake out mistakes like this? thanks, benson -- Jean-Baptiste Onofré jbono...@apache.org http://blog.nanthrax.net Talend - http://www.talend.com
Re: More entropy on bundle startup.
I don't think so, no. I do not do any dynamic installation. I use the Maven plugin to make an assembly with all the features I need. I then observe that the startup order is not deterministic from machine to machine, and it is particularly prone to change when I stop and start the container without clearing out the data directory. On Thu, Jul 14, 2016 at 1:39 AM, Jean-Baptiste Onofréwrote: > Hi Benson, > > I guess you are using the deploy folder, so the fileinstall deployer, right > ? > > Regards > JB > > > On 07/13/2016 10:56 PM, Benson Margulies wrote: >> >> Folks, >> >> We've had a couple of incidents of latent problems stemming from >> invalid assumptions on bundle start order. Everything seems to be >> fine, then some trivial change reveals that we've failed to ensure >> that service 'a' is available before component 'b' needs it. by and >> large, we use DS to get this right, but there are a few cases where it >> does not serve. >> >> I am wondering: is there some way to get _more_ randomness out of the >> startup process, to shake out mistakes like this? >> >> thanks, >> benson >> > > -- > Jean-Baptiste Onofré > jbono...@apache.org > http://blog.nanthrax.net > Talend - http://www.talend.com
Re: More entropy on bundle startup.
Hi Benson, I guess you are using the deploy folder, so the fileinstall deployer, right ? Regards JB On 07/13/2016 10:56 PM, Benson Margulies wrote: Folks, We've had a couple of incidents of latent problems stemming from invalid assumptions on bundle start order. Everything seems to be fine, then some trivial change reveals that we've failed to ensure that service 'a' is available before component 'b' needs it. by and large, we use DS to get this right, but there are a few cases where it does not serve. I am wondering: is there some way to get _more_ randomness out of the startup process, to shake out mistakes like this? thanks, benson -- Jean-Baptiste Onofré jbono...@apache.org http://blog.nanthrax.net Talend - http://www.talend.com
Re: More entropy on bundle startup.
I love this idea! We’ve been fighting the same issues as well. -Nick On 7/13/16, 4:56 PM, "Benson Margulies"wrote: >Folks, > >We've had a couple of incidents of latent problems stemming from >invalid assumptions on bundle start order. Everything seems to be >fine, then some trivial change reveals that we've failed to ensure >that service 'a' is available before component 'b' needs it. by and >large, we use DS to get this right, but there are a few cases where it >does not serve. > >I am wondering: is there some way to get _more_ randomness out of the >startup process, to shake out mistakes like this? > >thanks, >benson
More entropy on bundle startup.
Folks, We've had a couple of incidents of latent problems stemming from invalid assumptions on bundle start order. Everything seems to be fine, then some trivial change reveals that we've failed to ensure that service 'a' is available before component 'b' needs it. by and large, we use DS to get this right, but there are a few cases where it does not serve. I am wondering: is there some way to get _more_ randomness out of the startup process, to shake out mistakes like this? thanks, benson