From 89a982de555ececf45e1d2f9dc33c5e111f032ca Mon Sep 17 00:00:00 2001
From: Ruslan Kuprieiev <kupruser@gmail.com>
Date: Mon, 16 Dec 2019 13:34:11 +0200
Subject: [PATCH] document locking in `dvc run/repro`

Fixes #860
---
 static/docs/command-reference/repro.md | 38 ++++++++++++++++++++++++++
 static/docs/command-reference/run.md   |  8 ++++++
 2 files changed, 46 insertions(+)

diff --git a/static/docs/command-reference/repro.md b/static/docs/command-reference/repro.md
index a990c2b76c2..9b6baa2035f 100644
--- a/static/docs/command-reference/repro.md
+++ b/static/docs/command-reference/repro.md
@@ -45,6 +45,44 @@ files, intermediate or final results. It saves all the data files, intermediate
 or final results into the <abbr>DVC cache</abbr> (unless `--no-commit` option is
 specified), and updates stage files with the new checksum information.
 
+### Running other dvc commands in parallel
+
+See
+[Running other dvc commands in parallel](/doc/command-reference/run#running-other-dvc-commands-in-parallel).
+
+### Parallel stage execution
+
+Currently `dvc repro` is not able to parallelize execution by itself (see
+[iterative/dvc#755](https://github.com/iterative/dvc/issues/755)), so if you
+need to do that you could launch multiple `dvc repro`s yourself. For example,
+say your DAG looks something like:
+
+```
+$ dvc pipeline show --ascii result.py
++--------+          +--------+
+| A1.dvc |          | B1.dvc |
++--------+          +--------+
+     *                   *
+     *                   *
+     *                   *
++--------+          +--------+
+| A2.dvc |          | B2.dvc |
++--------+          +--------+
+          *         *
+           **     **
+             *   *
+        +------------+
+        | result.dvc |
+        +------------+
+```
+
+so it consists of two pipeline branches (pipeline `A` and pipeline `B`) and the
+final `result` stage. To reproduce both branches at the same time, you could run
+`dvc repro A2.dvc` and `dvc repro B2.dvc` at the same time (e.g. by running them
+in separate terminals). After both are done running, you could then run
+`dvc repro result.dvc` that will see that both branches are already up-to-date
+and will only run the final stage.
+
 ## Options
 
 - `-f`, `--force` - reproduce a pipeline, regenerating its results, even if no
diff --git a/static/docs/command-reference/run.md b/static/docs/command-reference/run.md
index 7c953794059..a399975dd68 100644
--- a/static/docs/command-reference/run.md
+++ b/static/docs/command-reference/run.md
@@ -52,6 +52,14 @@ captures data and <abbr>caches</abbr> relevant <abbr>data artifacts</abbr> along
 the way. See [this example](/doc/get-started/example-pipeline) to learn more and
 try creating a pipeline.
 
+### Running other dvc commands in parallel
+
+When running your command, DVC will remove the project lock (`.dvc/lock` file),
+so that you will be able to run other DVC commands in parallel. However, it uses
+per-path read-write locking instead, to guarantee that no two DVC instances
+would be writing to the same path and don't write to paths that are being read
+from by another instance.
+
 ### Avoiding unexpected behavior
 
 We don't want to tell you how to write your code! However, please be aware that