[
https://issues.apache.org/jira/browse/IMPALA-2839?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Armstrong resolved IMPALA-2839.
-----------------------------------
Resolution: Fixed
We've added a lot of debug or stress options over time, this probably isn't
useful to keep open.
> add stress options to exercise infrequently executed paths
> ----------------------------------------------------------
>
> Key: IMPALA-2839
> URL: https://issues.apache.org/jira/browse/IMPALA-2839
> Project: IMPALA
> Issue Type: Task
> Components: Backend
> Affects Versions: Impala 2.3.0
> Reporter: Daniel Hecht
> Priority: Minor
> Labels: ramp-up
>
> We should add debug-only stress options that help exercise more paths in our
> testing. We have one example already:
> {code:title=global-flags.cc}
> DEFINE_int32(stress_free_pool_alloc, 0, "A stress option which causes memory
> allocations "
> "to fail once every n allocations where n is the value of this flag.
> Effective in "
> "debug builds only.");
> {code}
> Some examples that we should add:
> * We could exercise this path by overriding the
> curr_tuple_pool_->total_allocated_bytes() > MAX_TUPLE_POOL_SIZE condition,
> which would might help test for IMPALA-2829:
> {code:title=AnalyticEvalNode::ProcessChildBatch()}
> // Transfer resources to prev_tuple_pool_ when enough resources have
> accumulated
> // and the prev_tuple_pool_ has already been transfered to an output batch.
> if (curr_tuple_pool_->total_allocated_bytes() > MAX_TUPLE_POOL_SIZE &&
> (prev_pool_last_result_idx_ == -1 || prev_pool_last_window_idx_ == -1))
> {
> prev_tuple_pool_->AcquireData(curr_tuple_pool_.get(), false);
> prev_pool_last_result_idx_ = last_result_idx_;
> if (window_tuples_.size() > 0) {
> prev_pool_last_window_idx_ = window_tuples_.back().first;
> } else {
> prev_pool_last_window_idx_ = -1;
> }
> VLOG_FILE << id() << " Transfer resources from curr to prev pool at idx: "
> << stream_idx << ", stores tuples with last result idx: "
> << prev_pool_last_result_idx_ << " last window idx: "
> << prev_pool_last_window_idx_;
> }
> {code}
> * We could cause RowBatch::AtCapacity() to become true more often than normal
> (will be easier to do after http://gerrit.cloudera.org:8080/#/c/1399/ is
> committed).
> Then we should enable these flags for some of our testing.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
