Syslog drain app error messages in app log stream#633
Syslog drain app error messages in app log stream#633corporatemax wants to merge 1 commit intocloudfoundry:mainfrom
Conversation
|
|
chombium
left a comment
There was a problem hiding this comment.
@corporatemax Thanks for the PR and the time invested in it. Generally it looks good. I have only few minor comments and I would wait for the other approvers to tell what they think.
|
|
||
| // WriteLog writes a message in the application log stream using a LogClient. | ||
| func (appLogEmitter *LoggregatorEmitter) WriteLog(appID string, message string) { | ||
| if appLogEmitter.logClient == nil || appID == "" { |
There was a problem hiding this comment.
I wonder if we should write a warning in the Syslog Agent's logs that the emitter is not set and the app dev won't know that something went wrong... On the other hand, I understand that this is only to safeguard the EmitLog call as everything is initialized properly in the syslog_agent.go cmd.
I feel like if we don't write a log message for the syslog agent, we are covering the misconfiguration... Of course, this might not be the write place as the log will be written on every WriteLog call.
@ctlong Any thoughts on this?
There was a problem hiding this comment.
I am not sure if the location of this file is sufficient as I am also calling it in the filtered binding fetcher. Should I move this in its separate package? If yes where exactly?
There was a problem hiding this comment.
Let's leave it as is for now until we think of something better.
There was a problem hiding this comment.
@chombium it sounds like you're worried that we do not have any mechanisms to prevent future misconfiguration of the AppLogEmitter when it's initialized in syslog_agent.go, is that right?
I think we could adequately cover this case with a new test in syslog_agent_test.go. That would seem like a valuable addition to me.
| "log" | ||
|
|
||
| "golang.org/x/net/context" | ||
| "log" |
There was a problem hiding this comment.
Was this reformatting? If so please run golangci-lint run ./... in the /src folder
There was a problem hiding this comment.
no this was goland - I ran the formatter on my side and commited the changes.
There was a problem hiding this comment.
This looks very strange to me. Usually, the standard libraries are grouped at the top... can you please double-check the formatting here, thanks
| drainDroppedMetric.Add(float64(missed)) | ||
|
|
||
| w.emitLoggregatorErrorLog(b.AppId, fmt.Sprintf("%d messages lost for application %s in user provided syslog drain with url %s", missed, b.AppId, anonymousUrl.String())) | ||
| w.loggregatorEmitter.WriteLog(b.AppId, fmt.Sprintf("%d messages lost for application %s in user provided syslog drain with url %s", missed, b.AppId, anonymousUrl.String())) |
There was a problem hiding this comment.
This is a system/operator message and we should adjust the text as we have a concrete App for which some logs are lost. IMO x messages lost for syslog drain with url y is enough. On the other hand if the same syslog drain url is used for multiple apps the message would be clearer this way. Though, the app id will be sent anyhow as label/tag of the log. I'm not totally sure if we should change the message.
@ctlong any thought on this?
There was a problem hiding this comment.
What is the final log message output to the user in this case? I think the log client, or some of the other logic in AppLogEmitter may prefix the message with some information. In any case, you're right, we should evaluate the full log envelope and see if anything should be tweaked to ensure that it is legible to an app developer.
bf6dd19 to
9f789db
Compare
chombium
left a comment
There was a problem hiding this comment.
@corporatemax Thanks for looking into the review and making adjustments. I'm happy with the how the things look like.
I wonder what @ctlong has to say about this PR.
|
@corporatemax Can you please check the unit tests? Do they fail because they run as GH Action or it's something else? |
|
@chombium the test states
I assume this test is flaky regarding the opened ports. |
|
@corporatemax I was able to reproduce the error locally on Ubuntu Linux 20.04 LTS when running both all the tests and only the failing test suite. All tests run: and running the single suite: At first I thought that we have some problem when all of the tests run together, but the problem is isolated, but it seems that the the problem is within the I've also tried running single test like: and that run properly. It's seems to me there are problems with how the tests are setup and run when running the whole |
I've rerun the tests as they run in the GitHub action and I could reproduce the problem. Therefore I will revoke my approval until all test are running properly
fb4fc55 to
7336dcf
Compare
chombium
left a comment
There was a problem hiding this comment.
Generally it looks ok, but I have some objections about the implementation as it differs slightly from the rest of the code.
| } | ||
|
|
||
| // AppLogEmitter abstracts the sending of a log to the application log stream. | ||
| type AppLogEmitter interface { |
There was a problem hiding this comment.
Do we have a need for an interface here? Who else implements the EmitLog method? Could this interface be deleted and the method attached to the DefaultAppLogEmitter?
| } | ||
|
|
||
| // DefaultAppLogEmitter is an implementation of AppLogEmitter which sends logs to an instance of a LogClient | ||
| type DefaultAppLogEmitter struct { |
There was a problem hiding this comment.
Can we rename this to something else please? The word default caught my eye...
Usually in the Loggregator's code we have a single struct, from which the default implementation is create and then we apply options... We should be consistent here as well with the naming and the implementation.
If the AppLogEmitter interface above is removed we have a good name for the struct.
8660c84 to
ca33637
Compare
|
@chombium I adressed the feedback regarding the DefaultAppLogEmitter. |
There was a problem hiding this comment.
Hi @corporatemax, I'm terribly sorry for keep you waiting that long. Thanks fro the changes so far. I have two more things which has to be taken care of. Ah, and please squash the commits.
| cfg Config, | ||
| m Metrics, | ||
| l *log.Logger, | ||
| factory syslog.AppLogEmitterFactory, |
There was a problem hiding this comment.
Can you be more specific in the naming? We already have a writerFactory down in the code, so it would be good if the variable naming is more concise.
| } | ||
| } | ||
|
|
||
| func NewDefaultAppLogEmitterFactory() DefaultAppLogEmitterFactory { |
There was a problem hiding this comment.
Same as above. Please remove "default" from the function name.
There was a problem hiding this comment.
removed the term Default from this function name.
78a1984 to
d36e467
Compare
4860796 to
3bc0a08
Compare
chombium
left a comment
There was a problem hiding this comment.
LGTM!
Thanks for the cooperation @corporatemax
|
@corporatemax Everything looks good now, but please fix the unit tests and the linting errors, before we merge this change |
3bc0a08 to
c4a5487
Compare
chombium
left a comment
There was a problem hiding this comment.
LGTM! Thanks for the quick fix @corporatemax
c4a5487 to
a45b295
Compare
No longer active in the project
🥺🥺🥺🥺🥺🥺
All the best Carson! ❤️
chombium
left a comment
There was a problem hiding this comment.
Hi @corporatemax, thanks for this iteration. I've found some unusual formatting and unchanged line moves. Can you please take a look at that?
There was a problem hiding this comment.
Hi @corporatemax ,
I've checked the PR and now we don't have problem with duplication as the error message is too deep in the TCPWriter. Now the problem is that the error is written only by connection problems. The thing with that is that most of the problems come from bad configuration. As we've already spoken, the validation happens in the Filtered Binding Fetcher and we should not spam the users. As we discussed previously, I've tried to find a way how to check if an App instance is running on a particular Diego Cell, so that we can emit the message from the Syslog Agent running on the cell. The only thing that I've found useful is the CAPI v3 /v3/apps/:guid/processes call. The problem with it, is that for it we would need an UAA user.
Today, we've discussed another more viable option with @fhambrec and checked if we can move the validation from the Syslog Agent's Filtered Binding Fetcher to the Syslog Binding Cache. The Cache runs only on scheduler VMs, so we won't have the spam multiplication problems caused by the number of VMs running the Syslog Agent and the number of application instances. We've concluded that if we move the Syslog Drain URL validation logic to the Get Method in the Binding Cache Poller, the logs would be emitted once per app every minute per Scheduler node. This is much less than previous as there is only one Scheduler VM per availability zone.
Can you please proceed with this new idea? I think you have everything in this PR, you will only need to decide if the validation logic hast to be moved to the binding cache and be done once and only the valid Syslog Drain Config gets into the cache.
Wdyt about this approach?
| } | ||
|
|
||
| func NewSyslogBindingCache(config Config, metrics Metrics, log *log.Logger) *SyslogBindingCache { | ||
| func NewSyslogBindingCache(config Config, metrics Metrics, l *log.Logger) *SyslogBindingCache { |
There was a problem hiding this comment.
please use log instead of l as the usage is far away down from the variable definition
There was a problem hiding this comment.
I changed it to logger as in this file the log package from the stdlib is imported and classhes with log symbols. Further I consistently named it logger in this file.
| func LoadConfig() Config { | ||
| cfg := Config{ | ||
| APIPollingInterval: 15 * time.Second, | ||
| APIPollingInterval: 60 * time.Second, |
There was a problem hiding this comment.
You don't need to change the defaults. Use the API_POLLING_INTERVAL envvar instead
There was a problem hiding this comment.
removed this change, was only for local testing. BTW I will also stash all my commits before merging.
baff1c5 to
090afa3
Compare
chombium
left a comment
There was a problem hiding this comment.
Hi @corporatemax,
thanks for the PR.
Generally it looks good, but the validation errors should land both in the platform and application logs and not only in app logs. Please take a look at my comments and think if it makes sense to move the app log emitter to another package as it is used both by the Syslog Agent and the Syslog Binding Cache.
src/pkg/egress/syslog/tcp.go
Outdated
| func (w *TCPWriter) connect() (net.Conn, error) { | ||
| conn, err := w.dialFunc(w.url.Host) | ||
| if err != nil { | ||
| logMessage := fmt.Sprintf("Failed to connect to %s", w.url.String()) |
There was a problem hiding this comment.
Should be written in the syslog agent's logs as well.
There was a problem hiding this comment.
should it be written with a logger (log.logger)? is sending it in the platform log stream sufficient?
There was a problem hiding this comment.
A simple log fmt.Sprintf as in the other places will be sufficient.
chombium
left a comment
There was a problem hiding this comment.
Please address my comments.
src/pkg/binding/poller.go
Outdated
| p.store.Set(bindings, bindingCount) | ||
| } | ||
|
|
||
| func checkBindings(bindings []Binding, emitter applog.LogEmitter, checker IPChecker, logger *log.Logger, failedHostsCache *simplecache.SimpleCache[string, bool], blacklistedDrainsGauge metrics.Gauge, invalidDrainsGauge metrics.Gauge) { |
There was a problem hiding this comment.
If the function is like this, the metrics will only be incremented inside the function.
As all of the parameters are already assigned to the poller, make the poller a receiver for this function and pass only the bindings. The function should also remove all the invalid bindings.
You can use pass the Bindings slice as a pointer an do the checks, everything else will be updated through the p. rerefences:
func (p *Poller) checkBindings(bindings *[]Binding)
| return net.IPv4(127, 0, 0, 1), nil | ||
| } | ||
|
|
||
| func (*dummyIPChecker) CheckBlacklist(ip net.IP) error { |
There was a problem hiding this comment.
Please add proper testing
src/pkg/egress/syslog/tcp.go
Outdated
| func (w *TCPWriter) connect() (net.Conn, error) { | ||
| conn, err := w.dialFunc(w.url.Host) | ||
| if err != nil { | ||
| logMessage := fmt.Sprintf("Failed to connect to %s", w.url.String()) |
There was a problem hiding this comment.
A simple log fmt.Sprintf as in the other places will be sufficient.
chombium
left a comment
There was a problem hiding this comment.
Please address the findings
8313b8b to
ed6a0a5
Compare
There was a problem hiding this comment.
Quite a big PR.. I found a bunch of things.
Comments prefixed with "MINOR" or "OPTIONAL" do not need to get addressed in my opinion.
Nice to see you improved separation of converns with a bunch of refactorings :) . Not an easy thing to tackle.
I also built a prototypical implementation for moving validation to a separate package and removing validation from binding fetcher at corporatemax#1
| "PPROF_PORT" => "#{p("metrics.pprof_port")}", | ||
| "USE_RFC3339" => "#{p("logging.format.timestamp") == "rfc3339"}", | ||
|
|
||
| "AGENT_CA_FILE_PATH" => "#{certs_dir}/agent_ca.crt", |
There was a problem hiding this comment.
MINOR: It is kind of confusing to have a different name for files like agent_ca.crt than in syslog-agent bmp.yml.erb. What is the reason for that?
| default: "deprecated" | ||
|
|
||
| agent.port: | ||
| description: "Port the agent is serving gRPC via mTLS" |
There was a problem hiding this comment.
MINOR: Also here. Agent is very generic as this repo contains multiple agents. I guess you mean the forwarder agent.
ed6a0a5 to
57f868d
Compare
|
I just noticed that poller might need some refactoring in a subsequent PR as it has quite a lot of if/else and loop statements. I drafted something in jorbaum@7e11944 just in case somebody wants to take a look how that could look like. |
57f868d to
2cf0297
Compare
2cf0297 to
2b8b7fb
Compare
2b8b7fb to
fdb70e0
Compare
Description
This change should resolve #579 .
I pulled the SyslogConnector.emitLoggregatorErrorLog function into the app_log_emitter.go file and moved its dependent variables accordingly and capsuled it with the AppLogEmitter struct.
Furthermore, I used the AppLogEmitter in the RetryWriter and the WriterFactory to provide error messages to the Application Developer who tries to configure a syslog drain but does not know why there are no logs showing up in the syslog drain target.
I did this also in the FilteredBindingFetcher as there missconfigurations leads to errors there.
Furthermore, to enable this feature, this PR in the cf-deployment repository is necessary, to distribute the TLS config for the forwarder-agent in the binding-cache cloudfoundry/cf-deployment#1301 .
Follow Ups:
Type of change
Testing performed?
Checklist:
mainbranch, or relevant version branchIf you have any questions, or want to get attention for a PR or issue please reach out on the #logging-and-metrics channel in the cloudfoundry slack