diff options
Diffstat (limited to 'lib')
-rw-r--r-- | lib/gitlab/git/repository.rb | 2 | ||||
-rw-r--r-- | lib/gitlab/git/tree.rb | 23 | ||||
-rw-r--r-- | lib/gitlab/git_access.rb | 2 | ||||
-rw-r--r-- | lib/gitlab/gitaly_client.rb | 23 | ||||
-rw-r--r-- | lib/gitlab/gitaly_client/commit_service.rb | 5 | ||||
-rw-r--r-- | lib/gitlab/job_waiter.rb | 8 | ||||
-rw-r--r-- | lib/gitlab/sidekiq_middleware/memory_killer.rb | 67 | ||||
-rw-r--r-- | lib/gitlab/sidekiq_middleware/shutdown.rb | 133 |
8 files changed, 187 insertions, 76 deletions
diff --git a/lib/gitlab/git/repository.rb b/lib/gitlab/git/repository.rb index e3cbf017e55..ddb9cf433eb 100644 --- a/lib/gitlab/git/repository.rb +++ b/lib/gitlab/git/repository.rb @@ -2206,7 +2206,7 @@ module Gitlab with_worktree(squash_path, branch, sparse_checkout_files: diff_files, env: env) do # Apply diff of the `diff_range` to the worktree diff = run_git!(%W(diff --binary #{diff_range})) - run_git!(%w(apply --index), chdir: squash_path, env: env) do |stdin| + run_git!(%w(apply --index --whitespace=nowarn), chdir: squash_path, env: env) do |stdin| stdin.binmode stdin.write(diff) end diff --git a/lib/gitlab/git/tree.rb b/lib/gitlab/git/tree.rb index ba6058fd3c9..b6ceb542dd1 100644 --- a/lib/gitlab/git/tree.rb +++ b/lib/gitlab/git/tree.rb @@ -14,14 +14,14 @@ module Gitlab # Uses rugged for raw objects # # Gitaly migration: https://gitlab.com/gitlab-org/gitaly/issues/320 - def where(repository, sha, path = nil) + def where(repository, sha, path = nil, recursive = false) path = nil if path == '' || path == '/' Gitlab::GitalyClient.migrate(:tree_entries) do |is_enabled| if is_enabled - repository.gitaly_commit_client.tree_entries(repository, sha, path) + repository.gitaly_commit_client.tree_entries(repository, sha, path, recursive) else - tree_entries_from_rugged(repository, sha, path) + tree_entries_from_rugged(repository, sha, path, recursive) end end end @@ -57,7 +57,22 @@ module Gitlab end end - def tree_entries_from_rugged(repository, sha, path) + def tree_entries_from_rugged(repository, sha, path, recursive) + current_path_entries = get_tree_entries_from_rugged(repository, sha, path) + ordered_entries = [] + + current_path_entries.each do |entry| + ordered_entries << entry + + if recursive && entry.dir? + ordered_entries.concat(tree_entries_from_rugged(repository, sha, entry.path, true)) + end + end + + ordered_entries + end + + def get_tree_entries_from_rugged(repository, sha, path) commit = repository.lookup(sha) root_tree = commit.tree diff --git a/lib/gitlab/git_access.rb b/lib/gitlab/git_access.rb index bbdb593d4e2..6400089a22f 100644 --- a/lib/gitlab/git_access.rb +++ b/lib/gitlab/git_access.rb @@ -199,7 +199,7 @@ module Gitlab def check_repository_existence! unless repository.exists? - raise UnauthorizedError, ERROR_MESSAGES[:no_repo] + raise NotFoundError, ERROR_MESSAGES[:no_repo] end end diff --git a/lib/gitlab/gitaly_client.rb b/lib/gitlab/gitaly_client.rb index c5d3e944f7d..9cd76630484 100644 --- a/lib/gitlab/gitaly_client.rb +++ b/lib/gitlab/gitaly_client.rb @@ -125,6 +125,8 @@ module Gitlab kwargs = yield(kwargs) if block_given? stub(service, storage).__send__(rpc, request, kwargs) # rubocop:disable GitlabSecurity/PublicSend + rescue GRPC::Unavailable => ex + handle_grpc_unavailable!(ex) ensure duration = Gitlab::Metrics::System.monotonic_time - start @@ -135,6 +137,27 @@ module Gitlab duration) end + def self.handle_grpc_unavailable!(ex) + status = ex.to_status + raise ex unless status.details == 'Endpoint read failed' + + # There is a bug in grpc 1.8.x that causes a client process to get stuck + # always raising '14:Endpoint read failed'. The only thing that we can + # do to recover is to restart the process. + # + # See https://gitlab.com/gitlab-org/gitaly/issues/1029 + + if Sidekiq.server? + raise Gitlab::SidekiqMiddleware::Shutdown::WantShutdown.new(ex.to_s) + else + # SIGQUIT requests a Unicorn worker to shut down gracefully after the current request. + Process.kill('QUIT', Process.pid) + end + + raise ex + end + private_class_method :handle_grpc_unavailable! + def self.current_transaction_labels Gitlab::Metrics::Transaction.current&.labels || {} end diff --git a/lib/gitlab/gitaly_client/commit_service.rb b/lib/gitlab/gitaly_client/commit_service.rb index 269a048cf5d..d60f57717b5 100644 --- a/lib/gitlab/gitaly_client/commit_service.rb +++ b/lib/gitlab/gitaly_client/commit_service.rb @@ -105,11 +105,12 @@ module Gitlab entry unless entry.oid.blank? end - def tree_entries(repository, revision, path) + def tree_entries(repository, revision, path, recursive) request = Gitaly::GetTreeEntriesRequest.new( repository: @gitaly_repo, revision: encode_binary(revision), - path: path.present? ? encode_binary(path) : '.' + path: path.present? ? encode_binary(path) : '.', + recursive: recursive ) response = GitalyClient.call(@repository.storage, :commit_service, :get_tree_entries, request, timeout: GitalyClient.medium_timeout) diff --git a/lib/gitlab/job_waiter.rb b/lib/gitlab/job_waiter.rb index f654508c391..f7a8eae0be4 100644 --- a/lib/gitlab/job_waiter.rb +++ b/lib/gitlab/job_waiter.rb @@ -15,16 +15,22 @@ module Gitlab # push to that array when done. Once the waiter has popped `count` items, it # knows all the jobs are done. class JobWaiter + KEY_PREFIX = "gitlab:job_waiter".freeze + def self.notify(key, jid) Gitlab::Redis::SharedState.with { |redis| redis.lpush(key, jid) } end + def self.key?(key) + key.is_a?(String) && key =~ /\A#{KEY_PREFIX}:\h{8}-\h{4}-\h{4}-\h{4}-\h{12}\z/ + end + attr_reader :key, :finished attr_accessor :jobs_remaining # jobs_remaining - the number of jobs left to wait for # key - The key of this waiter. - def initialize(jobs_remaining = 0, key = "gitlab:job_waiter:#{SecureRandom.uuid}") + def initialize(jobs_remaining = 0, key = "#{KEY_PREFIX}:#{SecureRandom.uuid}") @key = key @jobs_remaining = jobs_remaining @finished = [] diff --git a/lib/gitlab/sidekiq_middleware/memory_killer.rb b/lib/gitlab/sidekiq_middleware/memory_killer.rb deleted file mode 100644 index b89ae2505c9..00000000000 --- a/lib/gitlab/sidekiq_middleware/memory_killer.rb +++ /dev/null @@ -1,67 +0,0 @@ -module Gitlab - module SidekiqMiddleware - class MemoryKiller - # Default the RSS limit to 0, meaning the MemoryKiller is disabled - MAX_RSS = (ENV['SIDEKIQ_MEMORY_KILLER_MAX_RSS'] || 0).to_s.to_i - # Give Sidekiq 15 minutes of grace time after exceeding the RSS limit - GRACE_TIME = (ENV['SIDEKIQ_MEMORY_KILLER_GRACE_TIME'] || 15 * 60).to_s.to_i - # Wait 30 seconds for running jobs to finish during graceful shutdown - SHUTDOWN_WAIT = (ENV['SIDEKIQ_MEMORY_KILLER_SHUTDOWN_WAIT'] || 30).to_s.to_i - - # Create a mutex used to ensure there will be only one thread waiting to - # shut Sidekiq down - MUTEX = Mutex.new - - def call(worker, job, queue) - yield - - current_rss = get_rss - - return unless MAX_RSS > 0 && current_rss > MAX_RSS - - Thread.new do - # Return if another thread is already waiting to shut Sidekiq down - return unless MUTEX.try_lock - - Sidekiq.logger.warn "Sidekiq worker PID-#{pid} current RSS #{current_rss}"\ - " exceeds maximum RSS #{MAX_RSS} after finishing job #{worker.class} JID-#{job['jid']}" - Sidekiq.logger.warn "Sidekiq worker PID-#{pid} will stop fetching new jobs in #{GRACE_TIME} seconds, and will be shut down #{SHUTDOWN_WAIT} seconds later" - - # Wait `GRACE_TIME` to give the memory intensive job time to finish. - # Then, tell Sidekiq to stop fetching new jobs. - wait_and_signal(GRACE_TIME, 'SIGSTP', 'stop fetching new jobs') - - # Wait `SHUTDOWN_WAIT` to give already fetched jobs time to finish. - # Then, tell Sidekiq to gracefully shut down by giving jobs a few more - # moments to finish, killing and requeuing them if they didn't, and - # then terminating itself. - wait_and_signal(SHUTDOWN_WAIT, 'SIGTERM', 'gracefully shut down') - - # Wait for Sidekiq to shutdown gracefully, and kill it if it didn't. - wait_and_signal(Sidekiq.options[:timeout] + 2, 'SIGKILL', 'die') - end - end - - private - - def get_rss - output, status = Gitlab::Popen.popen(%W(ps -o rss= -p #{pid}), Rails.root.to_s) - return 0 unless status.zero? - - output.to_i - end - - def wait_and_signal(time, signal, explanation) - Sidekiq.logger.warn "waiting #{time} seconds before sending Sidekiq worker PID-#{pid} #{signal} (#{explanation})" - sleep(time) - - Sidekiq.logger.warn "sending Sidekiq worker PID-#{pid} #{signal} (#{explanation})" - Process.kill(signal, pid) - end - - def pid - Process.pid - end - end - end -end diff --git a/lib/gitlab/sidekiq_middleware/shutdown.rb b/lib/gitlab/sidekiq_middleware/shutdown.rb new file mode 100644 index 00000000000..c2b8d6de66e --- /dev/null +++ b/lib/gitlab/sidekiq_middleware/shutdown.rb @@ -0,0 +1,133 @@ +require 'mutex_m' + +module Gitlab + module SidekiqMiddleware + class Shutdown + extend Mutex_m + + # Default the RSS limit to 0, meaning the MemoryKiller is disabled + MAX_RSS = (ENV['SIDEKIQ_MEMORY_KILLER_MAX_RSS'] || 0).to_s.to_i + # Give Sidekiq 15 minutes of grace time after exceeding the RSS limit + GRACE_TIME = (ENV['SIDEKIQ_MEMORY_KILLER_GRACE_TIME'] || 15 * 60).to_s.to_i + # Wait 30 seconds for running jobs to finish during graceful shutdown + SHUTDOWN_WAIT = (ENV['SIDEKIQ_MEMORY_KILLER_SHUTDOWN_WAIT'] || 30).to_s.to_i + + # This exception can be used to request that the middleware start shutting down Sidekiq + WantShutdown = Class.new(StandardError) + + ShutdownWithoutRaise = Class.new(WantShutdown) + private_constant :ShutdownWithoutRaise + + # For testing only, to avoid race conditions (?) in Rspec mocks. + attr_reader :trace + + # We store the shutdown thread in a class variable to ensure that there + # can be only one shutdown thread in the process. + def self.create_shutdown_thread + mu_synchronize do + return unless @shutdown_thread.nil? + + @shutdown_thread = Thread.new { yield } + end + end + + # For testing only: so we can wait for the shutdown thread to finish. + def self.shutdown_thread + mu_synchronize { @shutdown_thread } + end + + # For testing only: so that we can reset the global state before each test. + def self.clear_shutdown_thread + mu_synchronize { @shutdown_thread = nil } + end + + def initialize + @trace = Queue.new if Rails.env.test? + end + + def call(worker, job, queue) + shutdown_exception = nil + + begin + yield + check_rss! + rescue WantShutdown => ex + shutdown_exception = ex + end + + return unless shutdown_exception + + self.class.create_shutdown_thread do + do_shutdown(worker, job, shutdown_exception) + end + + raise shutdown_exception unless shutdown_exception.is_a?(ShutdownWithoutRaise) + end + + private + + def do_shutdown(worker, job, shutdown_exception) + Sidekiq.logger.warn "Sidekiq worker PID-#{pid} shutting down because of #{shutdown_exception} after job "\ + "#{worker.class} JID-#{job['jid']}" + Sidekiq.logger.warn "Sidekiq worker PID-#{pid} will stop fetching new jobs in #{GRACE_TIME} seconds, and will be shut down #{SHUTDOWN_WAIT} seconds later" + + # Wait `GRACE_TIME` to give the memory intensive job time to finish. + # Then, tell Sidekiq to stop fetching new jobs. + wait_and_signal(GRACE_TIME, 'SIGTSTP', 'stop fetching new jobs') + + # Wait `SHUTDOWN_WAIT` to give already fetched jobs time to finish. + # Then, tell Sidekiq to gracefully shut down by giving jobs a few more + # moments to finish, killing and requeuing them if they didn't, and + # then terminating itself. + wait_and_signal(SHUTDOWN_WAIT, 'SIGTERM', 'gracefully shut down') + + # Wait for Sidekiq to shutdown gracefully, and kill it if it didn't. + wait_and_signal(Sidekiq.options[:timeout] + 2, 'SIGKILL', 'die') + end + + def check_rss! + return unless MAX_RSS > 0 + + current_rss = get_rss + return unless current_rss > MAX_RSS + + raise ShutdownWithoutRaise.new("current RSS #{current_rss} exceeds maximum RSS #{MAX_RSS}") + end + + def get_rss + output, status = Gitlab::Popen.popen(%W(ps -o rss= -p #{pid}), Rails.root.to_s) + return 0 unless status.zero? + + output.to_i + end + + def wait_and_signal(time, signal, explanation) + Sidekiq.logger.warn "waiting #{time} seconds before sending Sidekiq worker PID-#{pid} #{signal} (#{explanation})" + sleep(time) + + Sidekiq.logger.warn "sending Sidekiq worker PID-#{pid} #{signal} (#{explanation})" + kill(signal, pid) + end + + def pid + Process.pid + end + + def sleep(time) + if Rails.env.test? + @trace << [:sleep, time] + else + Kernel.sleep(time) + end + end + + def kill(signal, pid) + if Rails.env.test? + @trace << [:kill, signal, pid] + else + Process.kill(signal, pid) + end + end + end + end +end |