diff --git a/docs/issue-debugging-monitoring.md b/docs/issue-debugging-monitoring.md new file mode 100644 --- /dev/null +++ b/docs/issue-debugging-monitoring.md @@ -0,0 +1,125 @@ +# Tutorial: issue debugging and monitoring + +In order to debug issues happening in production, you need to get as much information as +possible on the issue. It helps to reproduce or directly fix the issue. In addition, you +want to monitor it to see how it evolves or if it is fixed for good. + +As a developer, the tools used for this at SWH are Sentry and Kibana. + +## Sentry overview + +URL: https://sentry.softwareheritage.org/ +Need a login password pair to access. No need VPN. + +To sign up, click "Request to join" and provide your SWH developer email address for the +admins to create the account. + +Official documentation: https://docs.sentry.io/product/ + +Sentry is specifically geared towards debugging production issues. In the "Issues" pane, +it presents issues grouped by similarity with statistics about their occurrence. Issues +can be filtered by project (here SWH VCS repository), environment, and time range. + +Viewing a particular issue, it presents the execution trace at the point of error, and +loads of contextual metadata about the running environment. You get the post-mortem +debugging information, including values of local variables, pretty-printed. The metadata +includes first and last occurrence, corresponding component versions, installed +packages, server environment. The breadcrumbs view shows several event log lines +produced in the same run prior to the error. These are not the logs produced by the +application, but events gathered through Sentry integrations. + +## Debugging SWH services with Sentry + +One difficulty may arise in search where an issue originates, due to the microservice +architecture used. An error of type `RemoteException` encapsulate an error occurring the +service called through a RPC call. If the information encapsulated in this error is not +sufficient, one would search for the complete trace of the original error by filtering +the issues by the project the originally erroring component is part of. + +Example: + +Sentry issue: `https://sentry.softwareheritage.org/organizations/swh/issues/5026/?project=11` +The error appear as `` +A request from a vault cooker to the storage service had a network error. + +Thanks to Sentry we see also which was the specific storage requested: + + `` + +Upon searching in the storage service issues, we find a corresponding `HttpResponseError`: +https://sentry.softwareheritage.org/organizations/swh/issues/3857/?project=3 + +We skip through the error reporting logic in the trace to get to the operation that was +performed. We see that this error comes from a RPC call to the objstorage service: + + HttpResponseError: "Download stream interrupted." at `swh/storage/objstorage.py` in `content_get` at line 41 + +This is a transient network error: it should not persist when retrying. So a solution +might be to add a retrying mechanism somewhere in this chain of RPC calls. + +## Issue monitoring with Sentry + +Aggregated error traces as shown in the "Issues" pane are the primary source of +information for monitoring. This includes the statistics of occurrence for a given +period of time. + +Sentry also comes with issue management features, that notably let you silence or +resolve errors. Silencing means the issue will still be recorded but not notified. +Resolving means the issue will be hidden from the default view, and any new occurrence +of it will specifically notify the issue owner that the issue still arises and is in +fact not resolved. Make sure an owner is associated to the issue, typically through +ownership rules set in the project settings. + +For more info on monitoring issues, refer to: +https://docs.sentry.io/product/error-monitoring/ + +## Kibana overview + +SWH wiki page: https://intranet.softwareheritage.org/wiki/Kibana + +URL: http://kibana0.internal.softwareheritage.org:5601/app/kibana +Need the SWH VPN but no credentials. + +Official documentation: https://www.elastic.co/guide/en/kibana/current/index.html + +Kibana is a vizualization UI for searching through indexed logs. You can search through +different sources of logs in the "Discover" pane. The sources configured include +application logs for SWH services and system logs. You can also access dashboards shared +by other on a particular topic or create our own from a saved search. + +There are 2 query languages which are quite similar: Lucene or KML. Whatever one you +choose, you will have the same querying capabilities. A query tries to match values for +specific keys, and support many predicates and combination of them. See the +documentation for KML: https://www.elastic.co/guide/en/kibana/current/kuery-query.html + +To get logs for a particular service, you have to know the name of its systemd unit and +the hostname of the production server providing this service. For a worker, switch the +index pattern to "swh_workers-*", for another SWH service switch it to "systemlogs-*". + +Example for getting swh-vault production logs: +In systemlogs-* index pattern: +`systemd_unit:"gunicorn-swh-vault.service" AND hostname:"vangogh"` + +Upon expanding a log entry with the leading arrow icon, you can inspect the entry in a +structured way. You can filter on particular values or fields, using the icons that are +left to the desired field. Fields including "message", "hostname" or "systemd_unit" are +often the most informational. You can also view the entry in context, several entries +before and after chronologically. + +## Issue monitoring with Kibana + +You can use Kibana saved searches and dashboards to follow issues based on associated +logs. Of course, we need to have logs produced that are related to the issue we want to +track. + +You can save a search, as opposed to only a query, to easily get back to it or include +it in a dashboard. Just click "Save" in the top toolbar above the search bar. It +includes the query, filters, selected columns, sorting and index pattern. + +Now you may want to have a customizable view of these logs, along with graphical +presentations. In the "Dashboard" pane, create a new dashboard. Click "add" in the top +toolbar and select your saved search. It will appear in resizeable panel. Now doing a +search will search trough the only data in the panels. + +To create more complete vizualizations including graphs, refer to: +https://www.elastic.co/guide/en/kibana/current/dashboard.html