[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ... #26

stephenc · 2015-04-24T09:37:12Z

...name collision avoidance

The current name collision avoidance uses the requests address, which could very likely be the same for all clients
as they could be being routed through a HTTP proxy (or two) so that is not a good disambiguator
We use a digest of the client's interfaces and MAC addresses and the remoteFSRoot to try and give a consistent ID
We ALWAYS append the ID if we have it as otherwise during reconnect the slaves with the same name will shuffle around
which defeats a lot of the login that Jenkins has internally based on slaves having a consistent name
In the event of legacy clients that do not have the ID we will let them connect with their name as long as there
is no online slave with that name. This does mean that where there are multiple legacy swarm clients with the
same name, only one can be on-line at any moment in time, but that is an improvement on the current where
once a shuffle starts, none can stay on-line

@reviewbybees

…or name collision avoidance - The current name collision avoidance uses the requests address, which could very likely be the same for all clients as they could be being routed through a HTTP proxy (or two) so that is not a good disambiguator - We use a digest of the client's interfaces and MAC addresses and the remoteFSRoot to try and give a consistent ID - We ALWAYS append the ID if we have it as otherwise during reconnect the slaves with the same name will shuffle around which defeats a lot of the login that Jenkins has internally based on slaves having a consistent name - In the event of legacy clients that do not have the ID we will let them connect with their name as long as there is no online slave with that name. This does mean that where there are multiple legacy swarm clients with the same name, only one can be on-line at any moment in time, but that is an improvement on the current where once a shuffle starts, none can stay on-line

…p growing

rsandell · 2015-04-24T15:07:40Z

client/src/main/java/hudson/plugins/swarm/SwarmClient.java

+    }
+
+    public static String getDigestOf(String text) {
+        try {


Apache commons has some nice util classes (org.apache.commons.codec.binary.Hex and
org.apache.commons.codec.digest.DigestUtils) that would make this a bit less bloated, and eliminate the need for toHexString

But that would bloat the swarm-client jar file

Ok, I assumed that the inclusion of org.apache.commons.httpclient would drag those with it as dependencies.

The client already has commons-codec as a dependency.

jtnord · 2015-04-24T16:28:06Z

client/src/main/java/hudson/plugins/swarm/SwarmClient.java

@@ -292,7 +307,9 @@ protected void createSwarmSlave(Candidate target) throws IOException, Interrupte
                + param("labels", labelStr)
                + param("toolLocations", toolLocationsStr)
                + "&secret=" + target.secret
-                + param("mode", options.mode.toUpperCase()));
+                + param("mode", options.mode.toUpperCase())


potential case issues in whacky locales... supply a defaultLocale.

jenkinsadmin · 2015-04-24T18:02:32Z

Thank you for a pull request! Please check this document for how the Jenkins project handles pull requests

stephenc · 2015-04-27T11:59:20Z

@jtnord @rsandell I think I have addressed your comments

jtnord · 2015-04-27T12:06:57Z

client/src/main/java/hudson/plugins/swarm/SwarmClient.java

+                    if (ia instanceof Inet4Address) {
+                        buf.append(ia.getHostAddress()).append('\n');
+                    } else if (ia instanceof Inet6Address) {
+                        buf.append(ia.getHostAddress()).append('\n');


Not sure why you don't want to include non IP adapters (it's just more entropy?) or collapse this into on if but not enough to fail the review.

jtnord · 2015-04-27T12:13:04Z

Whilst I'm sure that name won't contain characters that are not the same in UTF-8 and ISO 8859-1 as its a digest this is essentially lying.

stephenc · 2015-04-27T12:59:19Z

@jtnord should be fine now

jtnord · 2015-04-27T14:53:07Z

👍

mindjiver · 2015-04-27T15:11:45Z

@stephenc @jtnord so it this good enough to merge now?

stephenc · 2015-04-27T15:41:59Z

@mindjiver you are the maintainer of this plugin, so that is your call. I'm not going to ask you to merge it until either:

I get a second +1 from a bee;
12h has passed since my last commit and I have at least one +1

But that is our internal process which does not concern you as a maintainer of a Jenkins plugin. If you see a pull request that you are happy to merge, your plugin, your call.

mindjiver · 2015-04-27T18:39:50Z

Thanks for the clarification 😃 I'm happy to wait, so any problems with the 1.23 release should trickle in first.

christ66 · 2015-04-27T18:48:16Z

👍

stephenc · 2015-04-27T18:59:51Z

@mindjiver this pull request has now successfully completed the CloudBees internal processes and we are now asking you, as the maintainer of the plugin for any additional feedback with a view towards merging and cutting a new release with this fix.

[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ...

mindjiver · 2015-04-27T19:28:29Z

Lets go!

philbert · 2015-06-08T11:12:42Z

Can we make it configurable whether or not we want to append this randomised string to the slave name?

The premise of the PR works ok only if you are assuming something is not already being done to make sure the node name is unique (like using a hostname).

This change had broken all our jobs and made the node names long and ugly :(

stephenc · 2015-06-08T12:04:14Z

@Phiche the issue here is that a client with the same name connecting causes the whole thing to blow up.

You can have multiple swarm clients connecting from the same host and often cloned machines or VMs can end up with the same hostname.

Due to how the swarm clients connect, if we went with something like "first one to connect gets the name, all the others use the hash" you would still get slave rotation as A-1 may connect first initially but after the restart A-2 connects first and grabs the name without the hash

We cannot assume that a name passed on the CLI guarantees uniqueness either as the command to launch the swarm client may be part of the cloned image's startup scripts.

The workaround for you is to use an older build of the client as legacy clients will connect using their name.

I would view relying on nodes names for ephemeral nodes, such as those provided by swarm, is a bad plan and it would be better to use labels.

However, if you feel strongly towards the other position you could create a pull request and see what @mindjiver feels. My personal opinion is that what you ask is rather dangerous and if I were the maintainer of this project I would reject such a request. However I am not the maintainer, @mindjiver is, and the easiest way to see what he feels is to suggest a pull request for the feature you want and see what he says

stephenc · 2015-06-08T12:26:33Z

The simplest fix for your builds is just to add the old name to the --labels parameter when launching the client

If you do that, the builds will run on the correct nodes and the only remaining issue -for you- is the ugly display names.

philbert · 2015-06-08T12:40:45Z

Unfortunately the ugly display names is not just a trivial aesthetic. The names are quite meaningful in our case. They are carried the way all through the toolchain including host names, slave build names, jenkins job names, tags on EC2, and in Route53.

I liked having full control over the names even if jenkins dies in a fiery implosion if there was a name collision (which we never saw in hundreds of slaves being created/destroyed).

Downgrading the plugin is the preferred option at the moment.

mindjiver · 2015-07-21T18:29:46Z

I revisited this thread and I still think it's reasonable as it is. Hashes in node names is not very pretty but this was actually how we ended up doing it at my previous employer, so good that it's available straight in this plugin.

dawidmalina · 2015-07-29T11:30:03Z

I must agree with @Phiche that this option should be configurable. If you prefer to enable it by default it ok but please add ability to disable this feature. It forcing us to stay with 1.23 version.

If you have hostname collision why don't you run swarm in docker. We are doing it and it is working as a charm.

jenkinsci#26

stephenc added 2 commits April 24, 2015 10:36

Actually need to keep options.name as immutable or else it will kee…

0871658

…p growing

rsandell reviewed Apr 24, 2015
View reviewed changes

extra null safety

2132772

jtnord reviewed Apr 24, 2015
View reviewed changes

stephenc added 5 commits April 27, 2015 12:36

Add support for IPv6 addresses

7984ef2

Return key-value properties to support future extensions

640637c

Explain what's going wrong

78c8650

Use commons-codec DigestUtils

b7e1470

Specify locale

ff5dd53

jtnord reviewed Apr 27, 2015
View reviewed changes

oops missed the charset fix

96a1a84

mindjiver added a commit that referenced this pull request Apr 27, 2015

Merge pull request #26 from stephenc/jenkins-26558

5f622da

[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ...

mindjiver merged commit 5f622da into jenkinsci:master Apr 27, 2015

stephenc deleted the jenkins-26558 branch April 29, 2015 10:05

dawidmalina added a commit to dawidmalina/swarm-plugin that referenced this pull request Jul 29, 2015

Ability to disable unique ID for Clients

da0fd12

jenkinsci#26

dawidmalina mentioned this pull request Jul 29, 2015

Ability to disable unique ID for Clients #33

Merged

dawidmalina added a commit to dawidmalina/swarm-plugin that referenced this pull request Jul 29, 2015

Ability to disable unique ID for Clients

5552de1

jenkinsci#26

atarkowska mentioned this pull request Mar 24, 2016

Web build ome/devspace#19

Closed

jpmontez mentioned this pull request Aug 15, 2016

Appended UUIDs in Swarm node names causes mismatch in label lookups rcbops-qe/jenkins-oa#3

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ... #26

[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ... #26

stephenc commented Apr 24, 2015

rsandell Apr 24, 2015

stephenc Apr 24, 2015

rsandell Apr 24, 2015

rsandell Apr 24, 2015

jtnord Apr 24, 2015

jenkinsadmin commented Apr 24, 2015

stephenc commented Apr 27, 2015

jtnord Apr 27, 2015

jtnord commented Apr 27, 2015

stephenc commented Apr 27, 2015

jtnord commented Apr 27, 2015

mindjiver commented Apr 27, 2015

stephenc commented Apr 27, 2015

mindjiver commented Apr 27, 2015

christ66 commented Apr 27, 2015

stephenc commented Apr 27, 2015

mindjiver commented Apr 27, 2015

philbert commented Jun 8, 2015

stephenc commented Jun 8, 2015

stephenc commented Jun 8, 2015

philbert commented Jun 8, 2015

mindjiver commented Jul 21, 2015

dawidmalina commented Jul 29, 2015

[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ... #26

[FIXED JENKINS-26558] Clients should provide a unique ID to be used for ... #26

Conversation

stephenc commented Apr 24, 2015

rsandell Apr 24, 2015

Choose a reason for hiding this comment

stephenc Apr 24, 2015

Choose a reason for hiding this comment

rsandell Apr 24, 2015

Choose a reason for hiding this comment

rsandell Apr 24, 2015

Choose a reason for hiding this comment

jtnord Apr 24, 2015

Choose a reason for hiding this comment

jenkinsadmin commented Apr 24, 2015

stephenc commented Apr 27, 2015

jtnord Apr 27, 2015

Choose a reason for hiding this comment

jtnord commented Apr 27, 2015

stephenc commented Apr 27, 2015

jtnord commented Apr 27, 2015

mindjiver commented Apr 27, 2015

stephenc commented Apr 27, 2015

mindjiver commented Apr 27, 2015

christ66 commented Apr 27, 2015

stephenc commented Apr 27, 2015

mindjiver commented Apr 27, 2015

philbert commented Jun 8, 2015

stephenc commented Jun 8, 2015

stephenc commented Jun 8, 2015

philbert commented Jun 8, 2015

mindjiver commented Jul 21, 2015

dawidmalina commented Jul 29, 2015