RubyGems - gitlab-exporter - Versions diffs - 16.9.0 → 17.0.0 - Mend

gitlab-exporter 16.9.0 → 17.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (10) hide show

checksums.yaml +4 -4
data/Gemfile.lock +2 -18
data/config/gitlab-exporter.yml.example +0 -1
data/gitlab-exporter.gemspec +4 -2
data/lib/gitlab_exporter/database/bloat_btree.sql +143 -103
data/lib/gitlab_exporter/version.rb +1 -1
data/lib/gitlab_exporter/web_exporter.rb +96 -104
data/spec/web_exporter_spec.rb +127 -0
metadata +17 -30
data/lib/gitlab_exporter/rack_vulndb_255039_patch.rb +0 -35

checksums.yaml CHANGED Viewed

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 77bae663d5dff12133db411a27154cb55c100caf0bec1ecdd4dd4e8102c88994
-  data.tar.gz: d98c9128dc327d8757d4bb8d021cf863274f59588767ff877ce5ebe4c3274636
+  metadata.gz: 3cc0cadf1891c19066f388f384d51f47b9c489e8e0ee7c162565228fb661bada
+  data.tar.gz: 7ae7c0ac398b1832c58c2348297139fd88e3ff02e447669043990312160ed0a8
 SHA512:
-  metadata.gz: ff4b8dc86a0765ab98cf1e7ac97ee7016e09e1ec1e0a08db4366bdd9a66ae75dc34cb9d41ad5ed28668a6d279f641db780526ac3619b04b114f4c68aa29d8a74
-  data.tar.gz: a76a114f309436e969865d25cb3a0e7aa4b94d8239cb4d7f775958ef971d3a6747380d002adf8999d3a8be545ad3692980ae980dd5ed6e1a8533cc3f7bc87173
+  metadata.gz: 7a358dbc90fe7bed11fc33a179fe2eb726e0d218925a56e7f076cc4c51d14272de4adca57cd5ae084e0a57d51feed0cf331b2dbf3ae67a5c0a2a972c4dba1863
+  data.tar.gz: c937591c8d81ad5296955bd5a55f92bcdd6f2ecb0f1fe7cdd36e9f429e88f5bbf2da8bce8375e8897fec29eda18ea0dd60fcd1bbfc5070c49e9b886a0faf3037

data/Gemfile.lock CHANGED Viewed

@@ -1,17 +1,16 @@
 PATH
   remote: .
   specs:
-    gitlab-exporter (16.9.0)
+    gitlab-exporter (17.0.0)
+      base64 (= 0.2.0)
       connection_pool (= 2.5.5)
       deep_merge (~> 1.2.2)
       faraday (= 2.14.3)
       pg (= 1.6.3)
-      puma (= 8.0.1)
       quantile (= 0.2.1)
       redis (= 4.8.1)
       redis-namespace (= 1.11.0)
       sidekiq (= 6.5.12)
-      sinatra (~> 3.2.0)
       webrick (~> 1.9)
 GEM
@@ -31,23 +30,15 @@ GEM
     language_server-protocol (3.17.0.5)
     lint_roller (1.1.0)
     logger (1.7.0)
-    mustermann (3.0.3)
-      ruby2_keywords (~> 0.0.1)
-    nio4r (2.7.4)
     parallel (1.27.0)
     parser (3.3.9.0)
       ast (~> 2.4.1)
       racc
     pg (1.6.3)
     prism (1.9.0)
-    puma (8.0.1)
-      nio4r (~> 2.0)
     quantile (0.2.1)
     racc (1.8.1)
     rack (2.2.15)
-    rack-protection (3.2.0)
-      base64 (>= 0.1.0)
-      rack (~> 2.2, >= 2.2.4)
     rainbow (3.1.1)
     redis (4.8.1)
     redis-namespace (1.11.0)
@@ -81,17 +72,10 @@ GEM
       parser (>= 3.3.7.2)
       prism (~> 1.7)
     ruby-progressbar (1.13.0)
-    ruby2_keywords (0.0.5)
     sidekiq (6.5.12)
       connection_pool (>= 2.2.5, < 3)
       rack (~> 2.0)
       redis (>= 4.5.0, < 5)
-    sinatra (3.2.0)
-      mustermann (~> 3.0)
-      rack (~> 2.2, >= 2.2.4)
-      rack-protection (= 3.2.0)
-      tilt (~> 2.0)
-    tilt (2.6.0)
     unicode-display_width (3.1.4)
       unicode-emoji (~> 4.0, >= 4.0.4)
     unicode-emoji (4.0.4)

data/config/gitlab-exporter.yml.example CHANGED Viewed

@@ -6,7 +6,6 @@ db_common: &db_common
 # Web server config
 server:
-  name: webrick # cf. https://github.com/sinatra/sinatra#available-settings
   listen_address: 0.0.0.0
   listen_port: 9168
   # Maximum amount of memory to use in megabytes, after which the process is killed

data/gitlab-exporter.gemspec CHANGED Viewed

@@ -22,16 +22,18 @@ Gem::Specification.new do |s|
   s.required_ruby_version = Gem::Requirement.new(">= 3.0")
+  # base64 is required by sidekiq but not declared by it, and is no longer a
+  # default gem as of Ruby 3.4. It used to be pulled in transitively via
+  # sinatra -> rack-protection; declare it directly now that those are gone.
+  s.add_runtime_dependency "base64", "0.2.0"
   s.add_runtime_dependency "connection_pool", "2.5.5"
   s.add_runtime_dependency "deep_merge", "~> 1.2.2"
   s.add_runtime_dependency "faraday", "2.14.3"
   s.add_runtime_dependency "pg", "1.6.3"
-  s.add_runtime_dependency "puma", "8.0.1"
   s.add_runtime_dependency "quantile", "0.2.1"
   s.add_runtime_dependency "redis", "4.8.1"
   s.add_runtime_dependency "redis-namespace", "1.11.0"
   s.add_runtime_dependency "sidekiq", "6.5.12"
-  s.add_runtime_dependency "sinatra", "~> 3.2.0"
   s.add_runtime_dependency "webrick", "~> 1.9"
   s.add_development_dependency "rspec", "~> 3.12.0"

data/lib/gitlab_exporter/database/bloat_btree.sql CHANGED Viewed

@@ -1,107 +1,147 @@
 -- Originally from: https://github.com/ioguix/pgsql-bloat-estimation/blob/master/btree/btree_bloat.sql
--- WARNING: executed with a non-superuser role, the query inspect only index on tables you are granted to read.
+--
+-- Rewritten to use CROSS JOIN LATERAL stages so this query stays in sync with the
+-- `postgres_index_bloat_estimate(p_schema, p_idxname)` function used by the main GitLab
+-- application's reindexing pipeline:
+-- https://gitlab.com/gitlab-org/gitlab/-/blob/master/db/migrate/20260526120000_add_postgres_index_bloat_estimate_function.rb
+--
+-- Expanding the index columns inside a LATERAL join (instead of putting a set-returning
+-- function such as generate_series in the SELECT list) avoids the ProjectSet barrier that
+-- prevents the planner from pushing predicates down. See:
+-- https://gitlab.com/gitlab-org/gitlab/-/work_items/569363
+-- https://gitlab.com/gitlab-org/gitlab/-/work_items/599210
+--
+-- WARNING: executed with a non-superuser role, the query inspects only indexes on tables you are granted to read.
 -- WARNING: rows with is_na = 't' are known to have bad statistics ("name" type is not supported).
--- This query is compatible with PostgreSQL 12 and after
--- pg_stats is materialized so the planner sees the true row count before planning the join,
--- avoiding a catastrophic cardinality underestimate on PostgreSQL 17.
-WITH pg_stats_cached AS MATERIALIZED (
-  SELECT schemaname, tablename, attname, null_frac, avg_width
-  FROM pg_catalog.pg_stats
-)
-SELECT current_database(), nspname AS schemaname, tblname, idxname AS object_name, bs*(relpages)::bigint AS real_size,
-  bs*(relpages-est_pages)::bigint AS extra_size,
-  100 * (relpages-est_pages)::float / relpages AS extra_ratio,
-  fillfactor,
-  CASE WHEN relpages > est_pages_ff
-    THEN bs*(relpages-est_pages_ff)
+-- This query is compatible with PostgreSQL 9.3 and after (LATERAL joins).
+--
+-- Note: unlike the previous ioguix-style query, this form does NOT need to materialize
+-- pg_stats to stay fast on PostgreSQL 17. The old query joined pg_stats once as a single
+-- large join that the PG17 planner badly underestimated (see f18ac4b); here pg_stats is
+-- looked up per index inside the LATERAL, resolving through the pg_statistic indexes, so
+-- there is no single join for the planner to misestimate.
+SELECT
+  current_database(),
+  n.nspname AS schemaname,
+  ct.relname AS tblname,
+  ci.relname AS object_name,
+  c.bs * ci.relpages::bigint AS real_size,
+  c.bs * (ci.relpages - bloat.est_pages)::bigint AS extra_size,
+  100 * (ci.relpages - bloat.est_pages)::float / ci.relpages AS extra_ratio,
+  c.fillfactor,
+  CASE WHEN ci.relpages > bloat.est_pages_ff
+    THEN c.bs * (ci.relpages - bloat.est_pages_ff)
     ELSE 0
   END AS bloat_size,
-  100 * (relpages-est_pages_ff)::float / relpages AS bloat_ratio,
-  is_na
-  -- , 100-(pst).avg_leaf_density AS pst_avg_bloat, est_pages, index_tuple_hdr_bm, maxalign, pagehdr, nulldatawidth, nulldatahdrwidth, reltuples, relpages -- (DEBUG INFO)
-FROM (
-  SELECT coalesce(1 +
-         ceil(reltuples/floor((bs-pageopqdata-pagehdr)/(4+nulldatahdrwidth)::float)), 0 -- ItemIdData size + computed avg size of a tuple (nulldatahdrwidth)
-      ) AS est_pages,
-      coalesce(1 +
-         ceil(reltuples/floor((bs-pageopqdata-pagehdr)*fillfactor/(100*(4+nulldatahdrwidth)::float))), 0
-      ) AS est_pages_ff,
-      bs, nspname, tblname, idxname, relpages, fillfactor, is_na
-      -- , pgstatindex(idxoid) AS pst, index_tuple_hdr_bm, maxalign, pagehdr, nulldatawidth, nulldatahdrwidth, reltuples -- (DEBUG INFO)
-  FROM (
-      SELECT maxalign, bs, nspname, tblname, idxname, reltuples, relpages, idxoid, fillfactor,
-            ( index_tuple_hdr_bm +
-                maxalign - CASE -- Add padding to the index tuple header to align on MAXALIGN
-                  WHEN index_tuple_hdr_bm%maxalign = 0 THEN maxalign
-                  ELSE index_tuple_hdr_bm%maxalign
-                END
-              + nulldatawidth + maxalign - CASE -- Add padding to the data to align on MAXALIGN
-                  WHEN nulldatawidth = 0 THEN 0
-                  WHEN nulldatawidth::integer%maxalign = 0 THEN maxalign
-                  ELSE nulldatawidth::integer%maxalign
-                END
-            )::numeric AS nulldatahdrwidth, pagehdr, pageopqdata, is_na
-            -- , index_tuple_hdr_bm, nulldatawidth -- (DEBUG INFO)
-      FROM (
-          SELECT n.nspname, i.tblname, i.idxname, i.reltuples, i.relpages,
-              i.idxoid, i.fillfactor, current_setting('block_size')::numeric AS bs,
-              CASE -- MAXALIGN: 4 on 32bits, 8 on 64bits (and mingw32 ?)
-                WHEN version() ~ 'mingw32' OR version() ~ '64-bit|x86_64|ppc64|ia64|amd64' THEN 8
-                ELSE 4
-              END AS maxalign,
-              /* per page header, fixed size: 20 for 7.X, 24 for others */
-              24 AS pagehdr,
-              /* per page btree opaque data */
-              16 AS pageopqdata,
-              /* per tuple header: add IndexAttributeBitMapData if some cols are null-able */
-              CASE WHEN max(coalesce(s.null_frac,0)) = 0
-                  THEN 2 -- IndexTupleData size
-                  ELSE 2 + (( 32 + 8 - 1 ) / 8) -- IndexTupleData size + IndexAttributeBitMapData size ( max num filed per index + 8 - 1 /8)
-              END AS index_tuple_hdr_bm,
-              /* data len: we remove null values save space using it fractionnal part from stats */
-              sum( (1-coalesce(s.null_frac, 0)) * coalesce(s.avg_width, 1024)) AS nulldatawidth,
-              max( CASE WHEN i.atttypid = 'pg_catalog.name'::regtype THEN 1 ELSE 0 END ) > 0 AS is_na
-          FROM (
-              SELECT ct.relname AS tblname, ct.relnamespace, ic.idxname, ic.attpos, ic.indkey, ic.indkey[ic.attpos], ic.reltuples, ic.relpages, ic.tbloid, ic.idxoid, ic.fillfactor,
-                  coalesce(a1.attnum, a2.attnum) AS attnum, coalesce(a1.attname, a2.attname) AS attname, coalesce(a1.atttypid, a2.atttypid) AS atttypid,
-                  CASE WHEN a1.attnum IS NULL
-                  THEN ic.idxname
-                  ELSE ct.relname
-                  END AS attrelname
-              FROM (
-                  SELECT idxname, reltuples, relpages, tbloid, idxoid, fillfactor, indkey,
-                      pg_catalog.generate_series(1,indnatts) AS attpos
-                  FROM (
-                      SELECT ci.relname AS idxname, ci.reltuples, ci.relpages, i.indrelid AS tbloid,
-                          i.indexrelid AS idxoid,
-                          coalesce(substring(
-                              array_to_string(ci.reloptions, ' ')
-                              from 'fillfactor=([0-9]+)')::smallint, 90) AS fillfactor,
-                          i.indnatts,
-                          pg_catalog.string_to_array(pg_catalog.textin(
-                              pg_catalog.int2vectorout(i.indkey)),' ')::int[] AS indkey
-                      FROM pg_catalog.pg_index i
-                      JOIN pg_catalog.pg_class ci ON ci.oid = i.indexrelid
-                      WHERE ci.relam=(SELECT oid FROM pg_am WHERE amname = 'btree')
-                      AND ci.relpages > 0
-                  ) AS idx_data
-              ) AS ic
-              JOIN pg_catalog.pg_class ct ON ct.oid = ic.tbloid
-              LEFT JOIN pg_catalog.pg_attribute a1 ON
-                  ic.indkey[ic.attpos] <> 0
-                  AND a1.attrelid = ic.tbloid
-                  AND a1.attnum = ic.indkey[ic.attpos]
-              LEFT JOIN pg_catalog.pg_attribute a2 ON
-                  ic.indkey[ic.attpos] = 0
-                  AND a2.attrelid = ic.idxoid
-                  AND a2.attnum = ic.attpos
-            ) i
-            JOIN pg_catalog.pg_namespace n ON n.oid = i.relnamespace
-            JOIN pg_stats_cached s ON s.schemaname = n.nspname
-                                  AND s.tablename = i.attrelname
-                                  AND s.attname = i.attname
-            GROUP BY 1,2,3,4,5,6,7,8,9,10,11
-      ) AS rows_data_stats
-  ) AS rows_hdr_pdg_stats
-) AS relation_stats
-ORDER BY nspname, tblname, idxname;
+  100 * (ci.relpages - bloat.est_pages_ff)::float / ci.relpages AS bloat_ratio,
+  bloat.is_na
+FROM pg_catalog.pg_class ci
+JOIN pg_catalog.pg_index i     ON i.indexrelid = ci.oid
+JOIN pg_catalog.pg_class ct    ON ct.oid = i.indrelid
+JOIN pg_catalog.pg_namespace n ON n.oid = ct.relnamespace
+-- Per-index system constants and fillfactor.
+CROSS JOIN LATERAL (
+    SELECT
+        current_setting('block_size')::numeric AS bs,
+        CASE -- MAXALIGN: 4 on 32bits, 8 on 64bits (and mingw32 ?)
+            WHEN version() ~ 'mingw32'::text
+              OR version() ~ '64-bit|x86_64|ppc64|ia64|amd64'::text THEN 8
+            ELSE 4
+        END AS maxalign,
+        24 AS pagehdr,     -- per page header, fixed size: 24 for 8.3 and after
+        16 AS pageopqdata, -- per page btree opaque data
+        COALESCE(
+            substring(array_to_string(ci.reloptions, ' '::text) FROM 'fillfactor=([0-9]+)')::smallint::integer,
+            90
+        ) AS fillfactor
+) c
+-- Aggregate the per-column statistics for this index, expanding the index columns
+-- through a LATERAL generate_series rather than a SELECT-list set-returning function.
+CROSS JOIN LATERAL (
+    SELECT
+        max(CASE WHEN COALESCE(a1.atttypid, a2.atttypid) = 'pg_catalog.name'::regtype THEN 1 ELSE 0 END) > 0 AS is_na,
+        /* per tuple header: add IndexAttributeBitMapData if some cols are null-able */
+        CASE WHEN max(COALESCE(s_t.null_frac, s_x.null_frac, 0::real)) = 0::double precision
+            THEN 2 -- IndexTupleData size
+            ELSE 2 + (32 + 8 - 1) / 8 -- IndexTupleData size + IndexAttributeBitMapData size
+        END AS index_tuple_hdr_bm,
+        /* data len: we remove null values, saving space, using their fractional part from stats */
+        sum((1::double precision - COALESCE(s_t.null_frac, s_x.null_frac, 0::real))
+            * COALESCE(s_t.avg_width, s_x.avg_width, 1024)::double precision) AS nulldatawidth
+    FROM generate_series(1, i.indnatts::integer) AS gs(attpos)
+    -- Decode the indkey int2vector into an int[] once per index (not once per column).
+    CROSS JOIN LATERAL (
+        SELECT string_to_array(textin(int2vectorout(i.indkey)), ' '::text)::integer[] AS indkey_arr
+    ) ik
+    LEFT JOIN pg_catalog.pg_attribute a1
+           ON ik.indkey_arr[gs.attpos] <> 0
+          AND a1.attrelid = i.indrelid
+          AND a1.attnum   = ik.indkey_arr[gs.attpos]
+    LEFT JOIN pg_catalog.pg_attribute a2
+           ON ik.indkey_arr[gs.attpos] = 0
+          AND a2.attrelid = ci.oid
+          AND a2.attnum   = gs.attpos
+    LEFT JOIN pg_catalog.pg_stats s_t
+           ON s_t.schemaname = n.nspname
+          AND s_t.tablename  = ct.relname
+          AND s_t.attname    = a1.attname
+    LEFT JOIN pg_catalog.pg_stats s_x
+           ON s_x.schemaname = n.nspname
+          AND s_x.tablename  = ci.relname
+          AND s_x.attname    = a2.attname
+) agg
+-- Index tuple header width, aligned on MAXALIGN.
+CROSS JOIN LATERAL (
+    SELECT
+        (
+            (agg.index_tuple_hdr_bm + c.maxalign
+             - CASE -- Add padding to the index tuple header to align on MAXALIGN
+                 WHEN (agg.index_tuple_hdr_bm % c.maxalign) = 0 THEN c.maxalign
+                 ELSE agg.index_tuple_hdr_bm % c.maxalign
+               END)::double precision
+          + agg.nulldatawidth
+          + c.maxalign::double precision
+          - CASE -- Add padding to the data to align on MAXALIGN
+              WHEN agg.nulldatawidth = 0::double precision THEN 0
+              WHEN (agg.nulldatawidth::integer % c.maxalign) = 0 THEN c.maxalign
+              ELSE agg.nulldatawidth::integer % c.maxalign
+            END::double precision
+        )::numeric AS nulldatahdrwidth
+) hdr
+-- Estimated number of pages, with and without the fillfactor applied.
+CROSS JOIN LATERAL (
+    SELECT
+        agg.is_na,
+        -- NULLIF guards against division by zero: floor(...) reaches 0 once
+        -- nulldatahdrwidth exceeds the usable page size (an index whose columns are
+        -- entirely unanalyzed and fall back to the 1024-byte default width). Without
+        -- it, PostgreSQL raises a hard error that would fail the whole probe; instead
+        -- the NULL flows through the outer COALESCE to est_pages = 0.
+        COALESCE(
+            1::double precision + ceil(
+                ci.reltuples / NULLIF(floor(
+                    (c.bs - c.pageopqdata::numeric - c.pagehdr::numeric)::double precision
+                    / (4::numeric + hdr.nulldatahdrwidth)::double precision
+                ), 0)
+            ),
+            0::double precision
+        ) AS est_pages,
+        COALESCE(
+            1::double precision + ceil(
+                ci.reltuples / NULLIF(floor(
+                    ((c.bs - c.pageopqdata::numeric - c.pagehdr::numeric) * c.fillfactor::numeric)::double precision
+                    / (100::double precision * (4::numeric + hdr.nulldatahdrwidth)::double precision)
+                ), 0)
+            ),
+            0::double precision
+        ) AS est_pages_ff
+) bloat
+WHERE ci.relkind = 'i'
+  AND ci.relam = (SELECT oid FROM pg_am WHERE amname = 'btree'::name)
+  AND ci.relpages > 0
+ORDER BY n.nspname, ct.relname, ci.relname;

data/lib/gitlab_exporter/version.rb CHANGED Viewed

@@ -1,5 +1,5 @@
 module GitLab
   module Exporter
-    VERSION = "16.9.0".freeze
+    VERSION = "17.0.0".freeze
   end
 end

data/lib/gitlab_exporter/web_exporter.rb CHANGED Viewed

@@ -1,146 +1,138 @@
-require "sinatra/base"
+require "webrick"
 require "English"
 require "cgi"
+require "stringio"
-require_relative "rack_vulndb_255039_patch"
 require_relative "tls_helper"
 module GitLab
   module Exporter
-    # Metrics web exporter
-    class WebExporter < Sinatra::Base
-      # A middleware to kill the process if we exceeded a certain threshold
-      class MemoryKillerMiddleware
-        def initialize(app, memory_threshold)
-          @app = app
-          @memory_threshold = memory_threshold.to_i * 1024
-        end
-        def call(env)
-          if memory_usage > @memory_threshold
-            puts "Memory usage of #{memory_usage} exceeded threshold of #{@memory_threshold}, signalling KILL"
-            Process.kill("KILL", $PID)
-          end
-          @app.call(env)
-        end
-        private
-        def memory_usage
-          io = IO.popen(%W[ps -o rss= -p #{$PID}])
-          mem = io.read
-          io.close
-          return 0 unless $CHILD_STATUS.to_i.zero?
-          mem.to_i
-        end
-      end
-      # Performs a major GC after each request. We found that this helps to free up
-      # several MB of memory in conjunction with sricter malloc config.
-      # See https://gitlab.com/gitlab-org/gitlab/-/issues/297241
-      class RunGC
-        def initialize(app)
-          @app = app
-        end
-        def call(env)
-          @app.call(env).tap do
-            GC.start
-          end
-        end
-      end
+    # Metrics web exporter backed directly by WEBrick (no Sinatra/Rack).
+    class WebExporter
+      DEFAULT_WEB_SERVER = "webrick".freeze
+      DEFAULT_LISTEN_ADDRESS = "0.0.0.0".freeze
+      DEFAULT_LISTEN_PORT = 9168
+      DEFAULT_MEMORY_THRESHOLD = 1024
+      # Security response headers previously added by Sinatra's Rack::Protection
+      # middleware, which is gone now that we serve directly via WEBrick. Kept for
+      # parity so scanners and consumers see the same headers as before.
+      SECURITY_HEADERS = {
+        "X-Content-Type-Options" => "nosniff",
+        "X-Frame-Options" => "SAMEORIGIN",
+        "X-XSS-Protection" => "1; mode=block"
+      }.freeze
       class << self
         include TLSHelper
-        DEFAULT_WEB_SERVER = "webrick".freeze
         def setup(config)
-          setup_server(config[:server])
-          setup_probes(config[:probes])
+          server_config = config[:server] || {}
-          memory_threshold = (config[:server] && config[:server][:memory_threshold]) || 1024
-          use MemoryKillerMiddleware, memory_threshold
-          use Rack::Logger
-          use RunGC
+          memory_threshold = server_config.fetch(:memory_threshold, DEFAULT_MEMORY_THRESHOLD)
+          @memory_threshold = memory_threshold.to_i * 1024
+          @server = build_server(server_config)
+          mount_probes(config[:probes])
           # Defrag heap after everything is loaded into memory.
           GC.compact
         end
-        def logger
-          request.logger
+        def run!
+          trap_signals
+          @server.start
         end
-        def setup_server(config)
-          config ||= {}
+        private
-          set(:server, config.fetch(:name, DEFAULT_WEB_SERVER))
-          set(:port, config.fetch(:listen_port, 9168))
+        def build_server(config)
+          warn_deprecated_server(config[:name])
+          options = {
+            Port: config.fetch(:listen_port, DEFAULT_LISTEN_PORT),
+            BindAddress: config.fetch(:listen_address, DEFAULT_LISTEN_ADDRESS)
+          }
-          # Depending on whether TLS is enabled or not, bind string
-          # will be different.
           if config.fetch(:tls_enabled, "false").to_s == "true"
-            set_tls_config(config)
-          else
-            set(:bind, config.fetch(:listen_address, "0.0.0.0"))
+            validate_tls_config(config)
+            options.merge!(webrick_tls_config(config))
           end
+          WEBrick::HTTPServer.new(**options)
         end
-        def set_tls_config(config) # rubocop:disable Naming/AccessorMethodName
-          validate_tls_config(config)
+        # The web server used to be configurable (e.g. Puma) via the `server.name`
+        # setting when the exporter ran on top of Sinatra. It now always uses WEBrick,
+        # so the setting is ignored. Warn operators whose config still sets it.
+        #
+        # TODO: This is step one of a two-step deprecation. In a future major release,
+        # raise instead of warning so the ignored `server.name` setting fails loudly.
+        def warn_deprecated_server(name)
+          return if name.nil? || name.to_s == DEFAULT_WEB_SERVER
+          warn "DEPRECATION: `server.name` (#{name.inspect}) is no longer supported and is ignored; " \
+               "the exporter now always uses WEBrick. Please remove `server.name` from your config."
+        end
-          web_server = config.fetch(:name, DEFAULT_WEB_SERVER)
-          if web_server == "webrick"
-            set_webrick_tls(config)
-          elsif web_server == "puma"
-            set_puma_tls(config)
-          else
-            fail "TLS not supported for web server `#{web_server}`."
+        def mount_probes(probes)
+          (probes || {}).each do |probe_name, params|
+            opts =
+              if params.delete(:multiple)
+                params
+              else
+                { probe_name => params }
+              end
+            @server.mount_proc("/#{probe_name}") do |_request, response|
+              handle_probe(response, opts)
+            end
           end
         end
-        def set_webrick_tls(config) # rubocop:disable Naming/AccessorMethodName
-          server_settings = {}
-          server_settings.merge!(webrick_tls_config(config))
+        def handle_probe(response, opts)
+          # A memory killer that kills the process if we exceeded a certain threshold.
+          check_memory_threshold!
-          set(:bind, config.fetch(:listen_address, "0.0.0.0"))
-          set(:server_settings, server_settings)
-        end
+          prober = Prober.new(metrics: PrometheusMetrics.new(include_timestamp: false), logger: @server.logger, **opts)
+          prober.probe_all
+          buffer = StringIO.new
+          prober.write_to(buffer)
-        def set_puma_tls(config) # rubocop:disable Naming/AccessorMethodName
-          listen_address = config.fetch(:listen_address, "0.0.0.0")
-          listen_port = config.fetch(:listen_port, 8443)
-          tls_cert_path = CGI.escape(config.fetch(:tls_cert_path))
-          tls_key_path = CGI.escape(config.fetch(:tls_key_path))
+          SECURITY_HEADERS.each do |header, value|
+            response[header] = value
+          end
+          response.content_type = "text/plain; version=0.0.4; charset=utf-8"
+          response.body = buffer.string
+        ensure
+          # Performs a major GC after each request. We found that this helps to free up
+          # several MB of memory in conjunction with stricter malloc config.
+          # See https://gitlab.com/gitlab-org/gitlab/-/issues/297241
+          GC.start
+        end
-          bind_string = "ssl://#{listen_address}:#{listen_port}?cert=#{tls_cert_path}&key=#{tls_key_path}"
+        def check_memory_threshold!
+          usage = memory_usage
+          return if usage <= @memory_threshold
-          set(:bind, bind_string)
+          puts "Memory usage of #{usage} exceeded threshold of #{@memory_threshold}, signalling KILL"
+          Process.kill("KILL", $PID)
         end
-        def setup_probes(config)
-          (config || {}).each do |probe_name, params|
-            opts =
-              if params.delete(:multiple)
-                params
-              else
-                { probe_name => params }
-              end
+        def memory_usage
+          # Use the block form so the pipe is closed automatically even if the
+          # read raises, avoiding a file-descriptor leak.
+          mem = IO.popen(%W[ps -o rss= -p #{$PID}], &:read)
-            get "/#{probe_name}" do
-              content_type "text/plain; version=0.0.4; charset=utf-8"
-              prober = Prober.new(metrics: PrometheusMetrics.new(include_timestamp: false), logger: logger, **opts)
+          return 0 unless $CHILD_STATUS.to_i.zero?
-              prober.probe_all
-              prober.write_to(response)
+          mem.to_i
+        end
-              response
-            end
+        def trap_signals
+          %w[INT TERM].each do |signal|
+            trap(signal) { @server.shutdown }
           end
         end
       end

data/spec/web_exporter_spec.rb ADDED Viewed

@@ -0,0 +1,127 @@
+require "spec_helper"
+require "gitlab_exporter/web_exporter"
+require "net/http"
+require "timeout"
+describe GitLab::Exporter::WebExporter do
+  # Each test builds a fresh server via `.setup`; drop the reference afterwards
+  # so class-level state does not leak between examples.
+  after do
+    described_class.instance_variable_set(:@server, nil)
+  end
+  describe ".setup" do
+    let(:probes) { {} }
+    it "builds a WEBrick server with the configured port and bind address" do
+      options = capture_server_options(server: { listen_address: "127.0.0.1", listen_port: 12_345 })
+      expect(options).to include(Port: 12_345, BindAddress: "127.0.0.1")
+    end
+    it "falls back to the default port and bind address" do
+      options = capture_server_options({})
+      expect(options).to include(Port: 9168, BindAddress: "0.0.0.0")
+    end
+    it "mounts a handler for each configured probe" do
+      server = instance_double(WEBrick::HTTPServer)
+      allow(WEBrick::HTTPServer).to receive(:new).and_return(server)
+      expect(server).to receive(:mount_proc).with("/ruby")
+      described_class.setup(server: {}, probes: { ruby: { methods: %w[probe_gc], opts: {} } })
+    end
+    it "raises when TLS is enabled but no certificate is configured" do
+      expect {
+        described_class.setup(server: { tls_enabled: true }, probes: probes)
+      }.to raise_error(/tls_cert_path not specified/)
+    end
+    # Capture the keyword options passed to WEBrick without binding a real socket.
+    def capture_server_options(config)
+      captured = nil
+      allow(WEBrick::HTTPServer).to receive(:new) do |**options|
+        captured = options
+        instance_double(WEBrick::HTTPServer)
+      end
+      described_class.setup(config.merge(probes: {}))
+      captured
+    end
+  end
+  describe "deprecating the server.name setting" do
+    it "warns when a non-WEBrick server is requested" do
+      expect {
+        described_class.send(:warn_deprecated_server, "puma")
+      }.to output(/DEPRECATION.*no longer supported/).to_stderr
+    end
+    it "does not warn when the server is webrick" do
+      expect { described_class.send(:warn_deprecated_server, "webrick") }.not_to output.to_stderr
+    end
+    it "does not warn when no server name is set" do
+      expect { described_class.send(:warn_deprecated_server, nil) }.not_to output.to_stderr
+    end
+  end
+  describe "serving probe requests" do
+    let(:config) do
+      {
+        server: { listen_address: "127.0.0.1", listen_port: 0 },
+        probes: { ruby: { methods: %w[probe_gc], opts: {} } }
+      }
+    end
+    let(:server) { described_class.instance_variable_get(:@server) }
+    let(:base_uri) { "http://127.0.0.1:#{server.listeners.first.addr[1]}" }
+    before do
+      # WEBrick binds its logger and access log to $stderr at construction time
+      # and emits a startup banner; silence $stderr around setup so the log
+      # device is a throwaway buffer and the test output stays clean.
+      original_stderr = $stderr
+      $stderr = StringIO.new
+      begin
+        described_class.setup(config)
+      ensure
+        $stderr = original_stderr
+      end
+      @server_thread = Thread.new { server.start }
+      Timeout.timeout(5) { sleep 0.01 until server.status == :Running }
+    end
+    after do
+      server.shutdown
+      @server_thread&.join(5)
+    end
+    it "returns Prometheus metrics for a known probe" do
+      response = Net::HTTP.get_response(URI("#{base_uri}/ruby"))
+      expect(response.code).to eq("200")
+      expect(response.content_type).to eq("text/plain")
+      expect(response["Content-Type"]).to include("version=0.0.4")
+      expect(response.body).to match(/ruby_gc_stat_count \d+/)
+    end
+    it "sets the security headers previously provided by Rack::Protection" do
+      response = Net::HTTP.get_response(URI("#{base_uri}/ruby"))
+      expect(response["X-Content-Type-Options"]).to eq("nosniff")
+      expect(response["X-Frame-Options"]).to eq("SAMEORIGIN")
+      expect(response["X-XSS-Protection"]).to eq("1; mode=block")
+    end
+    it "returns 404 for unknown paths" do
+      response = Net::HTTP.get_response(URI("#{base_uri}/does-not-exist"))
+      expect(response.code).to eq("404")
+    end
+  end
+end

metadata CHANGED Viewed

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: gitlab-exporter
 version: !ruby/object:Gem::Version
-  version: 16.9.0
+  version: 17.0.0
 platform: ruby
 authors:
 - Pablo Carranza
@@ -10,6 +10,20 @@ bindir: bin
 cert_chain: []
 date: 2016-07-27 00:00:00.000000000 Z
 dependencies:
+- !ruby/object:Gem::Dependency
+  name: base64
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - '='
+      - !ruby/object:Gem::Version
+        version: 0.2.0
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - '='
+      - !ruby/object:Gem::Version
+        version: 0.2.0
 - !ruby/object:Gem::Dependency
   name: connection_pool
   requirement: !ruby/object:Gem::Requirement
@@ -66,20 +80,6 @@ dependencies:
     - - '='
       - !ruby/object:Gem::Version
         version: 1.6.3
-- !ruby/object:Gem::Dependency
-  name: puma
-  requirement: !ruby/object:Gem::Requirement
-    requirements:
-    - - '='
-      - !ruby/object:Gem::Version
-        version: 8.0.1
-  type: :runtime
-  prerelease: false
-  version_requirements: !ruby/object:Gem::Requirement
-    requirements:
-    - - '='
-      - !ruby/object:Gem::Version
-        version: 8.0.1
 - !ruby/object:Gem::Dependency
   name: quantile
   requirement: !ruby/object:Gem::Requirement
@@ -136,20 +136,6 @@ dependencies:
     - - '='
       - !ruby/object:Gem::Version
         version: 6.5.12
-- !ruby/object:Gem::Dependency
-  name: sinatra
-  requirement: !ruby/object:Gem::Requirement
-    requirements:
-    - - "~>"
-      - !ruby/object:Gem::Version
-        version: 3.2.0
-  type: :runtime
-  prerelease: false
-  version_requirements: !ruby/object:Gem::Requirement
-    requirements:
-    - - "~>"
-      - !ruby/object:Gem::Version
-        version: 3.2.0
 - !ruby/object:Gem::Dependency
   name: webrick
   requirement: !ruby/object:Gem::Requirement
@@ -233,7 +219,6 @@ files:
 - lib/gitlab_exporter/prober.rb
 - lib/gitlab_exporter/process.rb
 - lib/gitlab_exporter/prometheus.rb
-- lib/gitlab_exporter/rack_vulndb_255039_patch.rb
 - lib/gitlab_exporter/ruby.rb
 - lib/gitlab_exporter/sidekiq.rb
 - lib/gitlab_exporter/tls_helper.rb
@@ -258,6 +243,7 @@ files:
 - spec/sidekiq_spec.rb
 - spec/spec_helper.rb
 - spec/util_spec.rb
+- spec/web_exporter_spec.rb
 homepage: https://gitlab.com/gitlab-org/ruby/gems/gitlab-exporter
 licenses:
 - MIT
@@ -300,3 +286,4 @@ test_files:
 - spec/sidekiq_spec.rb
 - spec/spec_helper.rb
 - spec/util_spec.rb
+- spec/web_exporter_spec.rb

data/lib/gitlab_exporter/rack_vulndb_255039_patch.rb DELETED Viewed

@@ -1,35 +0,0 @@
-# frozen_string_literal: true
-if Gem.loaded_specs["rack"].version >= Gem::Version.new("3.0.0")
-  fail <<~ERR
-    This patch is unnecessary in Rack versions 3.0.0 or newer.
-    Please remove this file and the associated spec.
-    See https://github.com/rack/rack/blob/main/CHANGELOG.md#security (issue #1733)
-  ERR
-end
-# Patches a cache poisoning attack vector in Rack by not allowing semicolons
-# to delimit query parameters.
-# See https://github.com/rack/rack/issues/1732.
-#
-# Solution is taken from the same issue.
-#
-# The actual patch is due for release in Rack 3.0.0.
-module Rack
-  class Request # rubocop:disable Style/Documentation
-    Helpers.module_eval do
-      # rubocop: disable Naming/MethodName
-      def GET
-        if get_header(RACK_REQUEST_QUERY_STRING) == query_string
-          get_header(RACK_REQUEST_QUERY_HASH)
-        else
-          query_hash = parse_query(query_string, "&") # only allow ampersand here
-          set_header(RACK_REQUEST_QUERY_STRING, query_string)
-          set_header(RACK_REQUEST_QUERY_HASH, query_hash)
-        end
-      end
-      # rubocop: enable Naming/MethodName
-    end
-  end
-end