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.
- 28 Feb, 2022 1 commit
-
-
Michelle Nguyen authored
Summary: This diff adds the database tables which will be used to track a plugin's configuration. `data_retention_plugin_releases` will track the provider-specified configuration options for their plugin. For example, the preset scripts and config values that need to be filled out when enabling the plugin. This will be populated by a job, similar to artifact tracker. `org_data_retention_plugins` tracks an org's settings for a plugin, for example: their plugin version, what their values for each config setting are. This does not include a table for the longterm retention scripts that a user has yet. That will follow in a separate diff. Test Plan: Skaffold deploy, ensure that the migrations run. These tables are currently unused. Reviewers: vihang, nserrino Reviewed By: vihang JIRA Issues: PC-1437 Signed-off-by:
Michelle Nguyen <michellenguyen@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10855 GitOrigin-RevId: d47f1bdaf1ed30f0511bc8fbaf7e5b753ebafcd0
-
- 27 Feb, 2022 1 commit
-
-
Vihang Mehta authored
Summary: Upgrade to go 1.17 Test Plan: All tests pass on new docker image Reviewers: zasgar, oazizi, #third_party_approvers Reviewed By: zasgar, #third_party_approvers Signed-off-by:
Vihang Mehta <vihang@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10871 GitOrigin-RevId: a947abe327fdb87d016dbd1e20cf2d818a90b6da
-
- 26 Feb, 2022 4 commits
-
-
Vihang Mehta authored
Summary: Upgrade `github.com/nats-io/nats-server/v2` and `github.com/nats-io/nats-streaming-server` to address some CVEs. The rest of the upgrades were a side effect. Should address https://advisories.nats.io/CVE/CVE-2021-3127.txt https://advisories.nats.io/CVE/CVE-2022-24450.txt Test Plan: skaffold deploy of cloud + vizier countinue to work. Existing tests should also cover functionality. Reviewers: michelle, zasgar, #third_party_approvers Reviewed By: michelle, #third_party_approvers Signed-off-by:
-
Nick Lanam authored
Summary: I changed a diff before landing it, then landed it and a successor at the same time. I forgot to rebase the latter first, so changes were partially reverted. This fixes the final state. Test Plan: Run Cypress as usual. Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1429 Signed-off-by:
Nick Lanam <nlanam@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10868 GitOrigin-RevId: 7df81c43248b232737ddfd24fa9e931e67b06068
-
Nick Lanam authored
Summary: Tests the rest of the sidebar. Also stubs script execution for this test to speed up the test. Further diffs will refine the stubbing approach in general to keep most tests snappy. Test Plan: Run Cypress as normal (see `src/ui/cypress/README.md`). Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1428 Signed-off-by:
-
Nick Lanam authored
Summary: Tests that scripts execute, with basic verification of what actually ran. Test Plan: Run Cypress as normal (see `src/ui/cypress/README.md`). Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1424 Signed-off-by:
-
- 25 Feb, 2022 7 commits
-
-
Omid Azizi authored
Summary: To improve testability, always use the injected clock. This means removing any references to std::chrono::steady_clock::now() inside data_stream.cc. Also, improve the tests by checking for the disabled reason when a tracker is disabled. Finally, restore a disabled test. Test Plan: This is mostly test changes. Any changes in the clock semantics are still covered by the tests. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Subscribers: yzhao Signed-off-by:
Omid Azizi <oazizi@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10848 GitOrigin-RevId: 431bf81d370a10fa96866dbfea0e2785e50da1b5
-
yzhao1012 authored
Test Plan: Tested manually with stirling_wrapper Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Signed-off-by:
yzhao1012 <yzhao@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10856 GitOrigin-RevId: abd910ac0a8e9998c32021257cce1bc61fcd49d3
-
Omid Azizi authored
Summary: Augment some of our BPF event CONN_TRACE information to help with tracking. Test Plan: These are debug tools. Tested manually. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Signed-off-by:
-
Omid Azizi authored
Summary: Add some documentation to ConnTracker states. Test Plan: None required. This is just documentation. Reviewers: #stirling, jps, yzhao Reviewed By: #stirling, jps, yzhao Subscribers: yzhao Signed-off-by:
-
Omid Azizi authored
Summary: ConnID should be set as soon as the ConnTracker is created to avoid CONN_TRACEs with upid=0. Test Plan: Manually checked the CONN_TRACE logs. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Subscribers: yzhao Signed-off-by:
-
Omid Azizi authored
Summary: Tidying up some of the code for consistency. Also optimize to make fewer calls to std::chrono::now(). Test Plan: No functional changes. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Subscribers: yzhao Signed-off-by:
-
Omid Azizi authored
Summary: By exposing stirling_sources, we can control stirling sources with an environment variable. Test Plan: Manual Reviewers: #stirling, jps Reviewed By: #stirling, jps Subscribers: nserrino Signed-off-by:
-
- 24 Feb, 2022 16 commits
-
-
yzhao1012 authored
Test Plan: Tested manually and see the printed logs Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Signed-off-by:
-
Michelle Nguyen authored
Summary: We will track plugin releases similarly to how we track vizier/CLI releases using artifact tracker. Releases that are published on Github will be used to update the plugin releases table in the plugin service. Test Plan: Skaffold deploy, to ensure that service starts up and migration runs Reviewers: vihang Reviewed By: vihang JIRA Issues: PC-1437 Signed-off-by:
-
Michelle Nguyen authored
Summary: We are adding the plugin service to Pixie cloud. This is just boilerplate for launching the deployment/service on kubernetes. The service does not do anything yet. Test Plan: Skaffold deploy Reviewers: vihang Reviewed By: vihang JIRA Issues: PC-1438 Signed-off-by:
-
James Bartlett authored
Summary: Adds a UDTF that calls ReleaseFreeMemory on all the agents. This causes tcmalloc to release spans in the free list to the OS. This is useful for experimenting with memory usage. Test Plan: Checked that px._HeapStats, showed about 600MB of free spans in the page heap, then after running the UDTF, another call to `px._HeapStats` showed that 500MB of that was returned to the OS. Reviewers: oazizi, vihang, nserrino Reviewed By: vihang Signed-off-by:
James Bartlett <jamesbartlett@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10846 GitOrigin-RevId: 7acee822d2514fa562ea1895ab7bcee54e1b5f56
-
Vihang Mehta authored
Summary: `http.Error` sets headers but by the time `ResponseWriter.write` is called, headers are already sent. So we can't really do much with the error at that point. Just drop it instead. Test Plan: The sever shouldn't have logs saying ` http: superfluous response.WriteHeader call` anymore. Reviewers: jamesbartlett Reviewed By: jamesbartlett Signed-off-by:
-
Phillip Kuznetsov authored
Summary: gendoc.py generated python docs with an extra nesting. That meant the python docs wouldn't get generated properly. We must've given up on this earlier which is why we didn't notice the problem. Test Plan: Generated the docs and made sure the python docs render from the data. Reviewers: htroisi, michelle, vihang Reviewed By: vihang JIRA Issues: PP-3156 Signed-off-by:
Phillip Kuznetsov <pkuznetsov@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10840 GitOrigin-RevId: f5018c03aa42aacd349ae4bc3934eef66c30440a
-
Pete Stevenson authored
Summary: The Java symbolization agent responds to a "method unload" callback. Here, we add support on the Stirling side for the same. Essentially, this removes a symbol from our Java symbol index. Test Plan: Existing. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: yzhao, oazizi JIRA Issues: PP-2982, PP-3264 Signed-off-by:
Pete Stevenson <jps@pixielabs.ai> Differential Revision: https://phab.corp.pixielabs.ai/D10805 GitOrigin-RevId: 92db82567a1c91e608dbb507f10eafc3a4a53c7f
-
Pete Stevenson authored
Summary: Because of symbol unloading, keeping Java symbols cached is challenging. For now, make the Java symbolizer uncacheable. We do not anticipate significant loss of perf. because the Java symbolizer keeps its own symbol index in memory (i.e. is already like a symbol cache). Test Plan: Existing. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: yzhao, oazizi JIRA Issues: PP-2982, PP-3264 Signed-off-by:
-
Omid Azizi authored
Summary: Just a few typos and fixes noticed along the way. Test Plan: None. No functional changes. Reviewers: #stirling, jps Reviewed By: #stirling, jps Signed-off-by:
-
James Bartlett authored
Summary: Adds alternative DataStreamBuffer implementation, which maintains events in a sorted map, until a call to `Head()` at which point the first contiguous events are merged into a contiguous buffer. This comes at the expense of an extra copy, but makes `RemovePrefix()` and future eviction work much faster, and also has much better memory properties when there are "gaps" in the data. Note that this diff removes `Get()` as a public API of DataStreamBuffer, and changes all call sites to use `Head()` instead (all call sites were in tests). Also changes Head to be non-const. Also adds a benchmark for the two DataStreamBufferImpl's. Using the socket tracer benchmark, and google/benchmark's compare.py tool, the following are the p values for the null hypothesis that the old and the new implementation are the same. These data suggest that the new implementation is statistically significantly faster on all benchmarks: ``` BM_SocketTraceConnector/http1_no_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/http1_encode_chunked_no_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/mysql_no_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/postgres_no_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/cql_no_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/nats_no_gaps_pvalue 0.0013 0.0013 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/http1_inter_iter_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/mysql_inter_iter_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/postgres_inter_iter_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/nats_inter_iter_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/cql_inter_iter_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 BM_SocketTraceConnector/http1_intra_iter_gaps_pvalue 0.0002 0.0002 U Test, Repetitions: 10 vs 10 ``` The mean on those benchmarks for the new implementation: ``` BM_SocketTraceConnector/http1_no_gaps_mean 8.90ms 22MB BM_SocketTraceConnector/http1_encode_chunked_no_gaps_mean 23.17ms 23MB BM_SocketTraceConnector/mysql_no_gaps_mean 12.02ms 21MB BM_SocketTraceConnector/postgres_no_gaps_mean 16.25ms 23MB BM_SocketTraceConnector/cql_no_gaps_mean 8.20ms 20MB BM_SocketTraceConnector/nats_no_gaps_mean 45.20ms 23MB BM_SocketTraceConnector/http1_inter_iter_gaps_mean 39.08ms 23MB BM_SocketTraceConnector/mysql_inter_iter_gaps_mean 72.62ms 23MB BM_SocketTraceConnector/postgres_inter_iter_gaps_mean 99.01ms 25MB BM_SocketTraceConnector/nats_inter_iter_gaps_mean 219.77ms 24MB BM_SocketTraceConnector/cql_inter_iter_gaps_mean 37.61ms 22MB BM_SocketTraceConnector/http1_intra_iter_gaps_mean 36.24ms 64MB ``` and for the old implementation: ``` BM_SocketTraceConnector/http1_no_gaps_mean 14.91ms 36MB BM_SocketTraceConnector/http1_encode_chunked_no_gaps_mean 39.66ms 39MB BM_SocketTraceConnector/mysql_no_gaps_mean 19.47ms 38MB BM_SocketTraceConnector/postgres_no_gaps_mean 31.09ms 39MB BM_SocketTraceConnector/cql_no_gaps_mean 19.58ms 36MB BM_SocketTraceConnector/nats_no_gaps_mean 51.89ms 40MB BM_SocketTraceConnector/http1_inter_iter_gaps_mean 90.18ms 63MB BM_SocketTraceConnector/mysql_inter_iter_gaps_mean 124.34ms 62MB BM_SocketTraceConnector/postgres_inter_iter_gaps_mean 160.59ms 64MB BM_SocketTraceConnector/nats_inter_iter_gaps_mean 299.13ms 44MB BM_SocketTraceConnector/cql_inter_iter_gaps_mean 84.91ms 62MB BM_SocketTraceConnector/http1_intra_iter_gaps_mean 312.37ms 527MB ``` Particularly of note is the dramatic improvement in memory usage for the `intra_iter_gaps` case from the new to the old. Test Plan: - Tested performance of new implementation in benchmarks, see above. - Tested correctness by making the data_stream_buffer_test test both the new and the old implementation. - Also tested that a skaffold deploy with the new DataStreamBuffer works on my dev cluster. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: oazizi, yzhao JIRA Issues: PP-3223 Signed-off-by:
-
James Bartlett authored
Summary: Adds a `RecordGenerator` for NATS data to the socket tracer benchmark. Blocked by D10800. Test Plan: No production changes. Tested that all the input NATS records get output by the benchmark. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Subscribers: yzhao JIRA Issues: PP-3211 Signed-off-by:
-
James Bartlett authored
Summary: Adds a `RecordGenerator` for cql data to the socket tracer benchmark. Blocked by D10799 Test Plan: No production changes. Tested that all the input cql records get output by the benchmark. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Subscribers: yzhao JIRA Issues: PP-3211 Signed-off-by:
-
Vihang Mehta authored
Summary: D10804 broke the simple `wrk` based test. This fixes it. Test Plan: Used skaffold to deploy the server and client. Reviewers: jamesbartlett, nserrino, zasgar Reviewed By: jamesbartlett Signed-off-by:
-
Nick Lanam authored
Summary: Skips testing Ctrl/Cmd+Enter until script execution diff lands. Test Plan: Run Cypress as normal (see `src/ui/cypress/README.md`). Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1443 Signed-off-by:
-
Nick Lanam authored
Summary: Adding a simple button that does this: easy. Making it look okay, not overlap the content, be obviously for the details, etc: a bit harder. This makes the details pane a bit more complex so that it can do this: - Details pane has a sticky header - Subtle shadow on header only when not scrolled to the top - Close button is clearly marked to be for the details - Table state syncs with close button correctly - Light and dark mode both look right {F179767} Test Plan: Run any PxL script that yields tables. Expand a row. Scroll around in the details. Expand other rows, press the close button, expand different rows. Expansion state should always match between the row highlight and the expansion pane. Try with both overlay and inline scrollbars (Chrome default vs Firefox default respectively; both can be configured to do the opposite by the end user). Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1444 Signed-off-by: -
Nick Lanam authored
Summary: TSIA. Test Plan: Run local UI. Visit `/live`, `/live/clusters`, and `/live/clusters/`. All should redirect to `/live/clusters/DEFAULT_CLUSTER_ID` if authenticated. Reviewers: michelle, vihang Reviewed By: michelle JIRA Issues: PC-1425 Signed-off-by:
-
- 23 Feb, 2022 3 commits
-
-
Omid Azizi authored
Summary: Fix a seg-fault introduced in D10823. Test Plan: Manually deployed on to a cluster and left it running for a while. Previously, the seg-fault would trigger within minutes. This no longer happens. Reviewers: #stirling, michelle, yzhao Reviewed By: #stirling, yzhao Subscribers: yzhao Signed-off-by:
-
Phillip Kuznetsov authored
Summary: We used to use the opname for the docs. But when I was reading the docs, I found this to be immensely confusing. I switched over to using the PxL name instead. This change was used to make the output docs: https://github.com/pixie-io/docs.px.dev/pull/143 Test Plan: Rendered the output docs in local dev. Reviewers: vihang, michelle, nserrino Reviewed By: michelle Signed-off-by:
-
Michelle Nguyen authored
Summary: Our licenses generator wasn't able to pick these up, so adding them manually. Test Plan: Ran license generator Reviewers: vihang Reviewed By: vihang Signed-off-by:
-
- 22 Feb, 2022 5 commits
-
-
Omid Azizi authored
Summary: absl::bind_front has some advantages over std::bind. See https://abseil.io/tips/108 . C++20 introduces std::bind_front() too, so we can switch to that once we are using it. Test Plan: No functional changes - Existing tests. Reviewers: #stirling, yzhao Reviewed By: #stirling, yzhao Signed-off-by:
-
Omid Azizi authored
Summary: A config flag controls whether we scan for mmap events, which is in turn used to detect dynamically loaded libraries. When this flag is off, there could be a leak, since we add to the map but never drain it. In this diff, we make sure not to add any items to the map if the flag has disabled the feature. Test Plan: Rely on existing tests to ensure no functionality is broken. Manual check for the leak fix. Reviewers: #stirling, jamesbartlett Reviewed By: jamesbartlett Signed-off-by:
-
Omid Azizi authored
Summary: The custom decoder is faster by ~7% ``` ---------------------------------------------------------------- Benchmark Time CPU Iterations ---------------------------------------------------------------- BM_custom_body_parser 58485 ns 58480 ns 10275 BM_pico_body_parser 62858 ns 62841 ns 11012 ``` Test Plan: There's already test coverage for this. Just flipping the switch. Reviewers: #stirling, jamesbartlett, yzhao Reviewed By: #stirling, yzhao Signed-off-by:
-
Omid Azizi authored
Summary: In the rare case when a container process switches namespaces with a call to setns, NetworkStatsConnector can incorrectly produce stats from the wrong namespace. Mitigate this effect by choosing the oldest PID in the container. This provides stability in which PID is chosen for producing stats. We also assume here that the oldest PID is not the one switching namespaces, so a more robust fix may be required in the future. Note that this bug was found by monitoring the PEM itself. Stirling forks a Java agent attach process which then changes namespaces, causing mis-reported network stats. Test Plan: Test added for the ordering. Also checked manually that the mis-reported network stats are fixed. Reviewers: #stirling, michelle, jps Reviewed By: #stirling, jps Signed-off-by:
-
Omid Azizi authored
Summary: This is a test utility for limiting Stirling's monitoring to a smaller set of PIDs. Test Plan: This is test tooling, so rely on existing tests. Reviewers: #stirling, jps Reviewed By: #stirling, jps Subscribers: jps Signed-off-by:
-
- 21 Feb, 2022 1 commit
-
-
Omid Azizi authored
Summary: Add a test to ensure only the desired namepsace is affected. Test Plan: This is a test. Reviewers: #stirling, jps Reviewed By: #stirling, jps Subscribers: jps Signed-off-by:
-
- 18 Feb, 2022 2 commits
-
-
James Bartlett authored
Summary: Adds a `RecordGenerator` for mysql data. Blocked by D10798 Test Plan: No production changes. Tested that all the input mysql records get output by the benchmark. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi JIRA Issues: PP-3211 Signed-off-by:
-
Pete Stevenson authored
PerfProfiler: change args to java agent to indicate the symbol file path (vs. name stem, which was awkward). Summary: Previously, the symbolization agent args indicated a file name stem because the symbol file name was uniquified based on target process pid. The name was given as a name stem (in case the agent needed to create a different file name for the txt log file). This setup was awkward both because it was a name stem and because it required coordination of the symbol file name in several places. Now we uniquify based on injecting the target process pid into the artifacts path. In this diff, we change the JVMTI symbolization agent args to indicate only the artifacts path and place a standard symbol file name in the shared header file. Test Plan: Existing. Reviewers: #stirling, oazizi Reviewed By: #stirling, oazizi Subscribers: oazizi JIRA Issues: PP-2982 Signed-off-by:
-
