Mostly comments, one small refactor

Some actual arrow_build_table tests, a CI job

r 4.4 to 4.5

more imports, more test

More test

Attempt to use PPPM for binaries

More minimal CI job?

Also need working directory for renv?

Do actually need system deps

update renv

Call it cran?

is PPM actually the magic?

Author: jonkeane

.github/workflows/crossbow-nightly-report-r-tests.yml

@@ -0,0 +1,53 @@
+name: Crossbow nightly report R tests
+
+on:
+  push:
+    branches: [ main ]
+    paths:
+      - 'crossbow-nightly-report/**'
+  pull_request:
+    branches: [ main ]
+    paths:
+      - 'crossbow-nightly-report/**'
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ./crossbow-nightly-report
+
+    steps:
+    - uses: actions/checkout@v3
+
+
+    - name: Install system dependencies
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y libcurl4-openssl-dev libssl-dev libxml2-dev
+
+    - name: Set up R
+      uses: r-lib/actions/setup-r@v2
+      with:
+        r-version: '4.5.0'
+        use-public-rspm: true
+
+    - name: Restore packages using renv
+      uses: r-lib/actions/setup-renv@v2
+      with: 
+        working-directory: ./crossbow-nightly-report
+
+    - name: Install test dependencies
+      run: install.packages('testthat')
+      shell: Rscript {0}
+
+    - name: Run tests
+      run: |
+        library(testthat)
+        test_results <- test_dir("tests", reporter = "summary", stop_on_failure = FALSE)
+        test_result_df <- as.data.frame(test_results)
+        # Exit with error code if any tests failed
+        if (length(test_results) > 0 && any(c(test_result_df$failed, test_result_df$error))) {
+          quit(status = 1)
+        }
+      shell: Rscript {0}

.github/workflows/nightly_dashboard.yml

@@ -50,7 +50,7 @@ jobs:
 
       - uses: r-lib/actions/setup-r@v2
         with:
-          r-version: '4.4.0'
+          r-version: '4.5.0'
           use-public-rspm: true
 
       # Needed due to https://github.com/r-lib/actions/issues/618

.github/workflows/performance-release-report.yml

@@ -57,7 +57,7 @@ jobs:
       - name: Setup R
         uses: r-lib/actions/setup-r@v2
         with:
-          r-version: '4.4.0'
+          r-version: '4.5.0'
           use-public-rspm: true
 
       # Needed due to https://github.com/r-lib/actions/issues/618

crossbow-nightly-report/R/functions.R

@@ -1,3 +1,9 @@
+library(tibble)
+library(dplyr)
+library(lubridate)
+library(glue)
+library(tidyr)
+
 is_dev <- function() {
   Sys.getenv("GITHUB_ACTIONS") != "true"
 }
@@ -40,14 +46,24 @@ make_nice_names <- function(x) {
   toTitleCase(gsub("_", " ", names(x)))
 }
 
-arrow_build_table <- function(nightly_data, type, task) {
+get_commit <- function(df, label) {
+  df$arrow_commit[df$fail_label == label]
+}
+
+arrow_build_table <- function(nightly_data, type, task, to_day = today()) {
+  # Filter data for a specific build type and task
   type_task_data <- nightly_data %>%
     filter(build_type == type) %>%
     filter(task_name == task)
 
-  ## filter for when the most recent run is a failure
-  day_window <- today() - 2
+  # Look at yesterday's date to determine recent failures
+  # This is used as a window for identifying tasks that failed recently
+  day_window <- to_day - 1
+
+  # Get records where the task failed recently, order by date (newest first)
+  # and standardize task status values to "pass" and "fail"
   ordered_only_recent_fails <- type_task_data %>%
+    # Only keep records where the task name appears in yesterday's failures
     filter(
       task_name %in%
         task_name[nightly_date == day_window & task_status != "success"]
@@ -61,15 +77,22 @@ arrow_build_table <- function(nightly_data, type, task) {
       )
     )
 
+  # If there are no recent failures, return a success summary or a null summary if the task is not active
   if (nrow(ordered_only_recent_fails) == 0) {
-    ## if there are no failures, return a version of the table that reflects that
+    # Calculate days since the last run (regardless of status)
     days <- as.numeric(
       difftime(
-        ymd(Sys.Date(), tz = "UTC"),
+        ymd(to_day, tz = "UTC"),
         max(type_task_data$nightly_date)
       )
     )
+    # Create a summary with success information
     success_df <- type_task_data %>%
+      # Remove stale data by filtering out everything but the last ~2 days of runs
+      # this makes it so that jobs that have been deleted (but are still in the 120 day look back)
+      # don't continue to show up.
+      filter(nightly_date >= to_day - 2) %>%
+      # Then, take the most recent run since that's all we care about if there are no failures.
       slice_max(order_by = nightly_date) %>%
       mutate(
         since_last_successful_build = days,
@@ -87,33 +110,39 @@ arrow_build_table <- function(nightly_data, type, task) {
         last_successful_build,
         build_type
       )
+
     return(success_df)
   }
 
-  ## find first failure index
+  # Find the length of the most recent consecutive failure streak
+  # This uses run length encoding to identify the first sequence of failures
   idx_recent_fail <- rle(ordered_only_recent_fails$task_status)$lengths[1]
 
-  ## expand failure index and give it some names
+  # Create labels for the failure streak timeline
+  # This builds a dataframe with positions and labels for the recent failure sequence
   failure_df <- tibble(fails_plus_one = seq(1, idx_recent_fail + 1)) %>%
     mutate(
       fail_label = case_when(
-        fails_plus_one == idx_recent_fail ~ "first_failure",
-        fails_plus_one == 1 ~ "most_recent_failure",
-        fails_plus_one == idx_recent_fail + 1 ~ "last_successful_build",
-        TRUE ~ paste0(fails_plus_one, " days ago")
+        fails_plus_one == idx_recent_fail ~ "first_failure", # Where the failures began
+        fails_plus_one == 1 ~ "most_recent_failure", # The most recent failure
+        fails_plus_one == idx_recent_fail + 1 ~ "last_successful_build", # Last successful build before failures
+        TRUE ~ paste0(fails_plus_one, " days ago") # General failure timeline
       )
     ) %>%
+    # Only keep the most recent 9 days of failures or specific labeled events
     filter(fails_plus_one <= 9 | grepl("failure|build", fail_label))
 
-  ## inner_join to ordered data
+  # Join the failure timeline labels with the actual build data
   df <- ordered_only_recent_fails %>%
     rowid_to_column() %>%
     inner_join(failure_df, by = c("rowid" = "fails_plus_one"))
 
+  # Calculate days since last successful build
   if (all(type_task_data$task_status %in% "failure")) {
     days <- NA_real_
   } else {
-    ## days since last successful build (need to add one)
+    # Calculate days between most recent failure and last successful build
+    # Adding 1 to include the day of the failure
     days <- sum(
       as.numeric(
         difftime(
@@ -125,10 +154,7 @@ arrow_build_table <- function(nightly_data, type, task) {
     )
   }
 
-  get_commit <- function(label) {
-    df$arrow_commit[df$fail_label == label]
-  }
-
+  # Format the final result as a table with build status information (one row per task)
   df %>%
     arrange(desc(fail_label)) %>%
     mutate(
@@ -137,12 +163,14 @@ arrow_build_table <- function(nightly_data, type, task) {
       )
     ) %>%
     select(task_name, build_type, build_links, fail_label) %>%
+    # Reshape data to have one column for each failure stage
     pivot_wider(names_from = fail_label, values_from = build_links) %>%
+    # Add additional context columns
     mutate(
       since_last_successful_build = days,
       last_successful_commit = arrow_compare_links(
-        get_commit("last_successful_build"),
-        get_commit("first_failure")
+        get_commit(df, "last_successful_build"),
+        get_commit(df, "first_failure")
       ),
       most_recent_status = "failing",
       .after = build_type

crossbow-nightly-report/air.toml

@@ -0,0 +1,2 @@
+[format]
+line-width = 120