* overkill for most Atomics because they are usually irregularly
   * scattered in memory and thus don't interfere much with each
   * other. But Atomic objects residing in arrays will tend to be
   * placed adjacent to each other, and so will most often share
   * cache lines (with a huge negative performance impact) without
   * this precaution.
   *

   * 1] < [1, 2] < [2]}.
   *
   * <p>Note that {@code Collections.reverseOrder(lexicographical(comparator))} is not equivalent to
   * {@code lexicographical(Collections.reverseOrder(comparator))} (consider how each would order
   * {@code [1]} and {@code [1, 1]}).
   */
  // Note: 90% of the time we don't add type parameters or wildcards that serve only to "tweak" the

Jobs live in the [jobs.t](jobs.t) directory, these are defined in YML using a syntax 
simmilar to [JJB](https://elasticsearch-ci.elastic.co/view/Elasticsearch%20master/).
Macros are not allowed, and each job must be defined in its own file.
Merging of the default configuration is customized so unlike in standard JJB,
it recurses into YML objects. 
Further (internal) documentation  on the setup 

Source code in this repository is covered by (i) a dual license under the Server
Side Public License, v 1 and the Elastic License 2.0 or (ii) an Apache License
2.0 compatible license or (iii) solely under the Elastic License 2.0, in each
case, as noted in the applicable header. The default throughout the repository
is a dual license under the Server Side Public License, v 1 and the Elastic
License 2.0, unless the header specifies another license. Code that is licensed

 * <li>the unreleased <b>staged</b>, M.N-2.0 (N &gt; 2) on the `M.(N-2)` branch</li>
 * </ul>
 * <p>
 * Each build is only concerned with versions before it, as those are the ones that need to be tested
 * for backwards compatibility. We never look forward, and don't add forward facing version number to branches of previous
 * version.
 * <p>
 * Each branch has a current version, and expected compatible versions are parsed from the server code's Version` class.

import java.util.List;
import java.util.Map;
import org.jspecify.annotations.Nullable;

/**
 * Implementation of {@code Multimap} that uses an {@code ArrayList} to store the values for a given
 * key. A {@link HashMap} associates each key with an {@link ArrayList} of values.
 *
 * <p>When iterating through the collections supplied by this class, the ordering of values for a
 * given key agrees with the order in which the values were added.
 *

                session.getProjects().size());
        boolean parallel = nThreads > 1;
        // Propagate the parallel flag to the root session and all the cloned sessions in each project segment
        session.setParallel(parallel);
        for (ProjectSegment segment : projectBuilds) {
            segment.getSession().setParallel(parallel);
        }

- To be able to reliably use S3 multipart APIs feature of the SDKs without re-inventing the wheel of pre-signing the each URL in multipart API. This is very tedious to implement with all the scenarios of fault tolerance that's already implemented by the client SDK. The general client SDKs don't support multipart with presigned URLs.

chứ
chưa
chuyện
có
có thể
cứ
của
cùng
cũng
đã
đang
đây
để
đến nỗi
đều
điều
do
đó
được
dưới
gì
khi
không
là
lại
lên
lúc
mà
mỗi
một cách
này
nên
nếu
ngay
nhiều
như
nhưng
những
nơi
nữa
phải
qua
ra
rằng
rằng
rất
rất
rồi
sau
sẽ
so
sự
tại
theo
thì
trên
trước
từ
từng

    return CollectCollectors.toImmutableSetMultimap(keyFunction, valueFunction);
  }

  /**
   * Returns a {@code Collector} accumulating entries into an {@code ImmutableSetMultimap}. Each
   * input element is mapped to a key and a stream of values, each of which are put into the
   * resulting {@code Multimap}, in the encounter order of the stream and the encounter order of the
   * streams of values.
   *
   * <p>Example:
   *