This project is mirrored from https://gitee.com/cowcomic/pixie.git.
Pull mirroring failed .
Repository mirroring has been paused due to too many failed attempts. It can be resumed by a project maintainer.
Repository mirroring has been paused due to too many failed attempts. It can be resumed by a project maintainer.
- 12 Jan, 2022 5 commits
-
-
Frederick Ryckbosch authored
Summary: This should be more robust across MacOS and Linux. Closes #388 Fixes #306 Test Plan: Ran the scripts, ensured that the generated secrets look correct. Reviewers: zasgar, michelle Reviewed By: michelle Signed-off-by:
Frederick Ryckbosch <fryckbosch@newrelic.com> Differential Revision: https://phab.corp.pixielabs.ai/D10474 GitOrigin-RevId: 846d451eb0e70b9d4478bd389fbdab8925c5f00e
-
Zain Asgar authored
Summary: gperftools autogen.sh stage was creating files that were not hermetic. This lead to caching issues. This change should result in a much faster build. Test Plan: Jenkins Reviewers: vihang, michelle, oazizi, #third_party_approvers Reviewed By: vihang, #third_party_approvers Signed-off-by:
Zain Asgar <zasgar@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10473 GitOrigin-RevId: ac0c631262a8a86a953c73fb7bafc5ea536d1077
-
Nick Lanam authored
Summary: Achieved by rewriting the quantiles component. It was using a Vega chart before, which was like using a missile to cook a casserole. The features were/are: - Render a whisker chart without the left whisker (only p50, p90, and p99). - Color the percentile bars based on how "bad" they are (defined by the column using the chart) - Hover color and tooltip for percentile bars - Clicking a bar tells the data table to switch which percentile bar is selected and used for sorting. - Use a nonlinear (square root) scale. - Resize when the column does (this was not working before). This doesn't take a lot of code, and the component is used for a table cell that can appear dozens of times at once. Writing a more performant version made sense. Test Plan: - Run `px/cluster`. - Mess around with the latency column in the Services table: - The percentile bars in each cell have tooltips - The bars have hover states - The selected percentile (title of the column) is already highlighted - The color of the bars is green, orange, or red depending on how high the latency at that bar is - Resizing the column resizes the charts too - Clicking a percentile bar selects it (makes that the shown value to the right of the chart and changes the column title; changes which bar is highlighted in every row) - The chart should be much less performance intensive than before. Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1344 Signed-off-by:
Nick Lanam <nlanam@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10456 GitOrigin-RevId: aaea843c1b4422343a28b48cc04af8cff9f43ef5
-
Nick Lanam authored
Summary: Shows a little helper message under the autocomplete results when there were more hits than are shown. Uses the new `hasAdditionalMatches` property of the suggestions API. {F175940} Test Plan: Use the breadcrumbs, especially on `px/pod` args. When there are more than about five results - which happens in the `pod` arg - this implementation suggests that results aren't complete. The message doesn't appear when results are already comprehensive. Reviewers: michelle, vihang, philkuz Reviewed By: michelle JIRA Issues: PC-1293 Signed-off-by: -
Nick Lanam authored
Summary: TSIA. Test Plan: - Run `px/cluster`. - Sort the Services table by latency (the quantiles column) - Hide some columns - Resize some remaining columns - Click on a row to open it in the sidebar - Change which quantile is selected by clicking a pip in that column (if p50 is selected, try p90, etc) - Sorting and the latency column's header should update to use the selected percentile. Hidden columns, resized columns, and which row is selected should be preserved. Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1355 Signed-off-by:
-
- 11 Jan, 2022 18 commits
-
-
Vihang Mehta authored
Summary: If the redirectGet succeeds, then we redirect to `/auth/cli-auth-complete` In this scenario, we don't want to actually render the CodeBox, so don't set loading to false. Test Plan: yarn dev Reviewers: michelle, philkuz Reviewed By: michelle JIRA Issues: PC-1357 Signed-off-by:
Vihang Mehta <vihang@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10468 GitOrigin-RevId: 08ec8ca30c976b2119c13a999b2a893d35c23db1
-
Vihang Mehta authored
Summary: This couldn't be auto fetched. Test Plan: Ran the license generator Reviewers: michelle Reviewed By: michelle Signed-off-by:
-
Ryan Cheng authored
Summary: This diff adds the PxL script to visualize Kafka consumer rebalancing events, including an overview of the consumer groups and the stop-the-world delay. {F175655} Test Plan: Manually tested in the UI. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: oazizi Signed-off-by:Ryan Cheng <rcheng@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10437 GitOrigin-RevId: 5de8e6f63e32ba7b6b9719e630a09060f5bd7f33
-
Phillip Kuznetsov authored
Summary: We're tracking invite users requests, we should track reset invite links as well. Test Plan: tested on yarn dev, found data for it. Reviewers: nlanam, michelle, vihang Reviewed By: vihang Signed-off-by:
Phillip Kuznetsov <pkuznetsov@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10466 GitOrigin-RevId: 06872e6fd906491e354fa93d9b1928f2f5196f6e
-
Vihang Mehta authored
Summary: Our CLI auth callbacks used the just pass the auth access token to the CLI without checking anything. However the CLI doesn't like org less users. So now auth callbacks on the UI will always signup/login (previously we skipped UI login in cli mode), then check the org from the user creds, and redirect to the org setup page if the user doesn't have an org. The org setup page is told to rediect back to a CLI login complete page that renders the token or sends it to the cli server as appropriate. Test Plan: `px auth login` and `px auth login --manual` with an org less user now correctly send you to org setup and after org setup send the token to the cli. Checking the org from the cli auth.json ensures that the CLI successfully exchanged the access token for a valid user token. Reviewers: michelle, nserrino Reviewed By: michelle Subscribers: nlanam, philkuz Signed-off-by: -
Phillip Kuznetsov authored
Summary: Adding a simple analytics event for whenever a user clicks on the invite users button. I don't track extra data since there is none. Test Plan: Tested on yarn dev, didn't break when I opened up the modal. Reviewers: michelle, nlanam, vihang Reviewed By: nlanam JIRA Issues: PC-1350 Signed-off-by:
-
Phillip Kuznetsov authored
Summary: When you render the invite modal, React doesn't like that the text is null before the gql is done. Changing to an empty string to avoid that error and we get the same values. Got the following errors before: ``` Warning: `value` prop on `input` should not be null. Consider using an empty string to clear the component or `undefined` for uncontrolled components. ``` ``` Warning: A component is changing an uncontrolled input to be controlled. This is likely caused by the value changing from undefined to a defined value, which should not happen. Decide between using a controlled or uncontrolled input element for the lifetime of the component. More info: https://reactjs.org/link/controlled-components ``` Test Plan: yarn dev. before change, opening the modal threw a console error. After change, opening the modal worked without throwing errors. Reviewers: nlanam, michelle, vihang Reviewed By: nlanam Signed-off-by:
-
Natalie Serrino authored
Summary: We've noticed that certain clusters fail to send any indexer data to Elastic on Pixie Cloud. This persists even in the case where we restart the indexer in cloud or restart Vizier fon that cluster. We've also seen logs about stan subscription timeouts in the indexer. In addition, there are repeated errors of Viziers trying to reconnect to the indexer, even though the indexer supposedly already started listening to those Viziers. It seems the following might be happening, 1. Vizier tries to connect to the indexer to send over metadata 2. The metadata subscription fails due to timeout 3. Future attempts by that Vizier to connect to the indexer are blocked, because the indexer believes that the connection with that Vizier is already being handled (even though it's not because it failed to subscribe to stan). In this diff, the error from the stan subscription should now be propagated up and attributed to a particular Vizier so it's easier to diagnose if that's happening. In addition, if the indexer fails to establish a stan connection for a given Vizier, it will now allow subsequent retries for that Vizier rather than blocking them. This alone won't fix the full issue but it should hopefully address some of it and give us better insight into the cases of Viziers that seem unable to send metadata to the indexer. Test Plan: will want to run this in testing. Reviewers: michelle, philkuz, vihang Reviewed By: vihang JIRA Issues: PC-1354 Signed-off-by:
Natalie Serrino <nserrino@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10440 GitOrigin-RevId: 2e543d161431c75efe3fe2a85330d6368f8b4ab7
-
Michelle Nguyen authored
Summary: Currently, our autocomplete results return matches that are too fuzzy. After some investigation, it is because we are using the ngram analyzer for our search results. For ngrams, elastic typically recommends that the analyzer should only be used in the indexing phase. For the search phase, results will be less fuzzy if we use the standard analyzer. Also updated the index to clean up the un-used NS field, and removed some un-used tokenizers/analyzers/etc. This diff also revs the index. Test Plan: Deployed to testing cluster, and tested various inputs to see if they return reasonable results. Reviewers: vihang, nserrino Reviewed By: vihang JIRA Issues: PC-1348 Signed-off-by:
Michelle Nguyen <michellenguyen@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10453 GitOrigin-RevId: 2f784ae64701dbcc4fe24b190cdc9ef0cfcee6ae
-
Michelle Nguyen authored
Summary: We currently don't deduplicate results in autocomplete, so if there is statefulset which has restarted with a different UID, it will show up twice. instead, we only want to show the most recent statefulset. this diff uses elastic's collapse functionality to deduplicate results, and ranks the resource with the most recent updateVersion at the top. Test Plan: updated the suggester test to include two resources with the same name. ensure that the resource with the later updateVersion is returned in the results. Reviewers: nserrino, vihang Reviewed By: vihang JIRA Issues: PC-1308 Signed-off-by:
-
Michelle Nguyen authored
Summary: We recently updated cloudConn to add the clusterID to the pod annotations. However, this can be a nil map if the pod has no annotations, and we need to handle that case. Test Plan: deployed using Skaffold, with a cloudConnector without any preexisting annotations Reviewers: vihang Reviewed By: vihang Signed-off-by:
-
Dom Del Nano authored
Summary: This integrates the Mux protocol parser into the core of Stirling. The parser is still disabled by default. Test Plan: A BPF test is included, but is currently disabled. To be enabled when we flip the switch to turn this on. Reviewers: #stirling, rcheng Reviewed By: #stirling, rcheng Signed-off-by:
Dom Del Nano <ddelnano@gmail.com> Differential Revision: https://phab.corp.pixielabs.ai/D10458 GitOrigin-RevId: e1fd9e9de7b2266bb976770344cb87e2c44d410c
-
Vihang Mehta authored
Summary: When pixie is being deleted, the CRD might be deleted first and then the operator would remove the rest of pixie pods during the reconciliation phase. In such a scenario, the cloud conn reports the last status as degraded due to the missing CRD. This should prevent that from happening. Test Plan: skaffold Reviewers: michelle, nserrino Reviewed By: michelle JIRA Issues: PP-3106 Signed-off-by:
-
Phillip Kuznetsov authored
Summary: Use the identityProvider field as a property in User signup/login analytics. Test Plan: skaffold Reviewers: vihang, michelle Reviewed By: vihang JIRA Issues: PC-1351 Signed-off-by:
-
Phillip Kuznetsov authored
Summary: We want analytics for which IdentityProvider logins/signups come from. This exposes the info from the auth api. Test Plan: skaffold Reviewers: vihang, michelle Reviewed By: vihang Signed-off-by:
-
Vihang Mehta authored
Summary: We notice some clusters with operators reporting status messages that claim that they don't have an operator running. Add some logging to investigate why this might be the case. Test Plan: skaffold deploy vizier, remove the `viziers` scope from the clusterrole for cloud-conn. Notice the warning show up in logs. Reviewers: michelle, nserrino, philkuz Reviewed By: michelle JIRA Issues: PP-3106 Signed-off-by:
-
James Bartlett authored
Summary: This diff allows setting the total amount of data stored across all tables by the table store, as a flag/envvar on PEM. In the future, we will expose this to the user somehow but that part is not clear yet. This diff also allows setting how much of the total limit should be allocated to the http_events table. This is also a flag/envvar but there are no current plans to expose this further to the user. The default values are chosen such that the amount of data for each table is the same as prior to this diff as long as the flag/envvar is not set. Test Plan: Tried without setting envvar, and then ran `px._DebugTableInfo()` to see that the values are the same as prior to this diff. Changed the envvar and re-ran to make sure it changes the table max sizes. Reviewers: nserrino Reviewed By: nserrino JIRA Issues: PP-3170 Signed-off-by:
James Bartlett <jamesbartlett@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10449 GitOrigin-RevId: c005676d293b6fb26aa8aae62676d34b39b3f99b
-
Pete Stevenson authored
Summary: bpf_tools/bcc_symbolizer.cc: fix comments. Test Plan: None, this diff fixes comments. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Signed-off-by:
Pete Stevenson <jps@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10441 GitOrigin-RevId: d192f6bb698acac60896dc58f4c14f28cd1cd9c8
-
- 10 Jan, 2022 6 commits
-
-
Michelle Nguyen authored
Summary: When we have multiple, closely matching results in our autocomplete, oftentimes the terminated pods are ranked before the running resources. However, the user will most likely want to see stats/info on their running resources. This diff updates our elastic query to use a boosting query to rank similarly matching entities that are currently running higher than non-terminated. Originally, the issue was written that we should priotize more recent resources. However, some resources may be long-lived and have less recent updates, so I decided to boost just on state (which is also easier implementation-wise). Test Plan: Deploy to testing with a running cluster and play around with different inputs in autocomplete to see we have a good balance between ranking suggestions with closely matching names vs their state. Reviewers: nserrino, vihang Reviewed By: vihang JIRA Issues: PC-1294 Signed-off-by:
-
Nick Lanam authored
Summary: Switching clusters was hiding most of the page until the cluster finished loading. This looked like a refresh. With this change, components handle this intermediate state instead of hiding away until cluster info is ready. Several spots needed to handle this new case: - `ExecuteScriptButton` should be disabled. - `ScriptLoader`, which auto-runs scripts when the target / params change, needs to wait. - If not switching scripts, widget layout should be preserved. - Script execution needs to wait. - `ClusterLoadingComponent` needs to show a spinner. - To avoid flickering intermediate states, `ScriptLoader` and `LiveView` need to treat loading state for the cluster the same as waiting for the script. Test Plan: - Log out and back in to reset "seen" (cached) clusters. - Run `px/http_data`. - Switch to a cluster that hasn't been seen in this session. Layout should be preserved. Run button should be disabled while waiting for cluster information to load, and stay that way while the script is running. Script should run against new cluster. - Switch back to the first cluster. Same thing should happen. - Switch to `px/tcp_retransmits` (or any other mutation script). It should not auto-run. - Run the mutation script. - Switch clusters. After a brief loading spinner to connect to the new cluster, the script should NOT auto-run and a message should appear as before. - Switch back again. Same thing should happen. Basically, the only thing that should be different is that switching clusters doesn't flash the entire page with `Loading...` and should instead preserve the page like switching args in a script does. Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1346 Signed-off-by:
-
Phillip Kuznetsov authored
Summary: We want to enable analytics for: 1. logins/signups with invite code 2. CreateOrg calls vs auto-created orgs from Gsuite signups Test Plan: Skaffolded Reviewers: vihang Reviewed By: vihang JIRA Issues: PC-1351 Signed-off-by:
-
Michelle Nguyen authored
Summary: Ever since we removed reading/writing to the NS field in our index, there was a one-off error here. Things will also look more correct once we rev our index, since there is old indexed data which does write to the NS field in elastic (these will all go away). Test Plan: Deploy to testing with a new index, look at the highlighting and confirm it actually looks as expected. Reviewers: nserrino, vihang Reviewed By: vihang JIRA Issues: PC-1347 Signed-off-by:
-
Vihang Mehta authored
Summary: This field is duplicated in UserInfo, never set and accidentally read once. Remove it so we don't accidentally try to use this unused field. Test Plan: Existing build and test. Reviewers: philkuz, nserrino, michelle Reviewed By: philkuz Signed-off-by:
-
Michelle Nguyen authored
Summary: We need to increase our number of cloud-proxy pods to handle higher loads Test Plan: bumped up # of replicas manually Reviewers: zasgar, vihang Reviewed By: vihang Signed-off-by:
-
- 08 Jan, 2022 2 commits
-
-
Pete Stevenson authored
In the java symbolization agent, use a run-time arg. to inject the path prefix for the symbol log file. Incorporate the same into the unit test; fixes the issue of colliding test cases. Summary: In the java symbolization agent, use a run-time arg. to inject the path prefix for the symbol log file. Incorporate the same into the unit test; fixes the issue of colliding test cases when multiple users run the test on the same host simultaneously. Test Plan: Symbolization agent unit test. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: yzhao, oazizi Signed-off-by:
-
Phillip Kuznetsov authored
Summary: Adds a python script that generates the api docs for the python api Test Plan: Tested and matches the old generated docs that we lost somewhere Reviewers: vihang, #third_party_approvers Reviewed By: vihang, #third_party_approvers Signed-off-by:
-
- 07 Jan, 2022 3 commits
-
-
Phillip Kuznetsov authored
Summary: I keep running make gazelle and hpack keeps generating a bazel file. I feel like we should just check one in so that we don't have stray files. Test Plan: n/a Reviewers: yzhao, vihang, #third_party_approvers Reviewed By: vihang, #third_party_approvers Signed-off-by:
-
Nick Lanam authored
Summary: Fixes two minor issues. - Running a mutation script manually didn't clear the manual flag afterward. - Switching clusters wasn't clearing script results in this scenario (this issue was hidden because the script was re-running instead). Aside: the first fix I tried removed the `manual` flag in favor of a different way to do the same thing. However, fixing the bugs associated with the flag turned out to be much cleaner. Test Plan: Run `px/tcp_drops`, manually, on a cluster in the live view. Then switch clusters. Now the script should not run on the new cluster. Additionally, it shouldn't run if you refresh either. Pressing `Run` or using the keyboard shortcut will still correctly run the script. Reviewers: michelle, vihang, philkuz Reviewed By: philkuz JIRA Issues: PC-1204 Signed-off-by:
-
Ryan Cheng authored
Summary: This diff adds full body parsing for the SyncGroup opcode. Test Plan: Unit tests and existing bpf test. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: oazizi Signed-off-by:
-
- 06 Jan, 2022 6 commits
-
-
Natalie Serrino authored
PC-1293: Autocomplete Field API should include information about whether or not additional results exist Summary: We want to add some more context to Live View UI script breadcrumbs using the Autocomplete service AutocompleteField API. It should say whether or not more results exist for a given query. This diff adds in a new boolean, HasAdditionalMatches, which is true when there are more matches available for a given search query. That way, if the user sees 5 matches, they will know if that is the complete set or if more exist. The UI has also been updated to use the new response format, but the UI doesn't currently consume this added boolean. Test Plan: added a unit test, tested dev cloud with dev UI. Reviewers: michelle, philkuz, nlanam Reviewed By: michelle, nlanam JIRA Issues: PC-1293 Signed-off-by:
-
Nick Lanam authored
Summary: Does a few things: - Uses dirty state of the PxL and Vis editors, rather than editor open state, to detect changes. This was PC-1024, and was one cause of PC-692. - Makes saving the editor also reconcile args. For example, removing a variable from the vis spec removes it from the retained args. The breadcrumbs were doing this, but the other update logic wasn't. - Tweaks logic for reconciliation: if a variable's default value was an empty string before, and becomes a non-empty string, it now fills it in with the new default value instead of leaving it empty. Test Plan: Run `px/http_data`. Remove the `source_filter` variable from the vis spec. Change the script to not need that variable. Close the editor. Run again. Updates should be reflected immediately. For another scenario, start from `px/http_data` again and add a default value for `source_filter`. Run again. The new default will be used right away. Reviewers: michelle, vihang, philkuz R...
-
Nick Lanam authored
Summary: Not a real fix, but this at least gives a savvy user a bit more info than "your vis spec is wrong" when the real problem was that the script refers to a variable that isn't in the vis spec. Split from another unrelated diff. Test Plan: Run `px/http_data`. Remove the `source_filter` variable from the vis spec. Run again. You get the same not-so-helpful snackbar, but the real error goes to the browser console. Reviewers: michelle, vihang Reviewed By: michelle Subscribers: philkuz Signed-off-by:
-
Nick Lanam authored
Summary: It was already a no-op, but it was still appearing in command help. This does not remove the component; that still exists albeit as dead code. Test Plan: Open Live UI. Use `Shift`+`/` to open keyboard shortcut help. Note there is no longer an entry for the Command Input; the old shortcut (`Ctrl/Cmd`+`K`) is no longer consumed by the UI. Instead, the default browser behavior happens (in Chrome and Firefox, it jumps to the URL bar in search mode). Reviewers: michelle, vihang, philkuz Reviewed By: philkuz JIRA Issues: PC-1343 Signed-off-by:
-
Vihang Mehta authored
Summary: We don't respect their field and throw an error, so let's remove it. Test Plan: Existing tests all pass. Reviewers: michelle, zasgar Reviewed By: michelle Signed-off-by:
-
Vihang Mehta authored
Summary: We are in the process of deprecating direct mode. As such, this change lets user toggle from direct mode to passthrough mode but disallows the inverse. Test Plan: Added some unit tests Reviewers: michelle, zasgar Reviewed By: michelle Signed-off-by:
-
