Add simplified critical path scheduler to improve build times - #2177
Merged
Conversation
The existing algorithm doesn't work because it strictly requires that all outputs are visited before updating an edge. So any task downstream from a task with multiple out-edges may get ignored. The fix is to always propagate your critical time to the next input node, and only place it in the queue if you offer a higher critical time.
1. Move EdgePriorityQueue to graph.h and inherit from priority_queue 2. Add comment about edge->critical_time()
AddTarget cannot add edges to the ready queue before the critical time has been computed.
|
I would have preferred the approach from #2019. The number of edges is a quite inaccurate estimate for the build time of the critical path. Also, I feel the mentioned problem1 is an orthogonal problem that should be solved differently, e.g. with pools. Both are symptoms that heavily depend on the particular project and do not necessarily generalize to other projects. Have you considered only prioritizing the critical path (as if with infinite parallelism), and schedule other edges in parallel to it according to other priorities, such as minimizing peak resource usage2? Footnotes
|
The historic build time is inaccurate, too, if e.g., ccache is used. |
4 tasks
13 tasks
elliotgoodrich
added a commit
to elliotgoodrich/trimja
that referenced
this pull request
Mar 23, 2025
If a build is going to fail, we want it to fail as fast as possible. This will give quicker feedback to the developer and reduce build resources. To facilitate this, we want to prioritize build edges that depend on changed files over other build edges that are at the same dependency level. e.g. for C++ projects we want to start compiling modified source files before other sources files that belong to the same library/executable. In this case we shouldn't increase (on average) the total build time on a successful fun, but an unsuccessful build would fail much faster. In ninja 1.12.0 a critical path schedule was added ninja-build/ninja#2177, which prioritizes build commands based on their distance from the build target. If two or more build commands have the same distance, it looks like the one that appears first in the build file is built first. Before ninja 1.12.0, the one that appears first in the build file is prioritized. This commit floats all affected commands to the top of the ninja build file (as far as it is possible at least) so that older ninja versions will work, and it helps ninja 1.12.0 in the case of ties. A future change will need to be made to artificially lengthen the critical path for affected files so that ninja 1.12.0+ will be fully supported.
elliotgoodrich
added a commit
to elliotgoodrich/trimja
that referenced
this pull request
Mar 23, 2025
If a build is going to fail, we want it to fail as fast as possible. This will give quicker feedback to the developer and reduce build resources. To facilitate this, we want to prioritize build edges that depend on changed files over other build edges that are at the same dependency level. e.g. for C++ projects we want to start compiling modified source files before other sources files that belong to the same library/executable. In this case we shouldn't increase (on average) the total build time on a successful fun, but an unsuccessful build would fail much faster. In ninja 1.12.0 a critical path schedule was added ninja-build/ninja#2177, which prioritizes build commands based on their distance from the build target. If two or more build commands have the same distance, it looks like the one that appears first in the build file is built first. Before ninja 1.12.0, the one that appears first in the build file is prioritized. This commit floats all affected commands to the top of the ninja build file (as far as it is possible at least) so that older ninja versions will work, and it helps ninja 1.12.0 in the case of ties. A future change will need to be made to artificially lengthen the critical path for affected files so that ninja 1.12.0+ will be fully supported.
elliotgoodrich
added a commit
to elliotgoodrich/trimja
that referenced
this pull request
Mar 23, 2025
If a build is going to fail, we want it to fail as fast as possible. This will give quicker feedback to the developer and reduce build resources. To facilitate this, we want to prioritize build edges that depend on changed files over other build edges that are at the same dependency level. e.g. for C++ projects we want to start compiling modified source files before other sources files that belong to the same library/executable. In this case we shouldn't increase (on average) the total build time on a successful fun, but an unsuccessful build would fail much faster. In ninja 1.12.0 a critical path schedule was added ninja-build/ninja#2177, which prioritizes build commands based on their distance from the build target. If two or more build commands have the same distance, it looks like the one that appears first in the build file is built first. Before ninja 1.12.0, the one that appears first in the build file is prioritized. This commit floats all affected commands to the top of the ninja build file (as far as it is possible at least) so that older ninja versions will work, and it helps ninja 1.12.0 in the case of ties. A future change will need to be made to artificially lengthen the critical path for affected files so that ninja 1.12.0+ will be fully supported.
elliotgoodrich
added a commit
to elliotgoodrich/trimja
that referenced
this pull request
Mar 23, 2025
If a build is going to fail, we want it to fail as fast as possible. This will give quicker feedback to the developer and reduce build resources. To facilitate this, we want to prioritize build edges that depend on changed files over other build edges that are at the same dependency level. e.g. for C++ projects we want to start compiling modified source files before other sources files that belong to the same library/executable. In this case we shouldn't increase (on average) the total build time on a successful fun, but an unsuccessful build would fail much faster. In ninja 1.12.0 a critical path schedule was added ninja-build/ninja#2177, which prioritizes build commands based on their distance from the build target. If two or more build commands have the same distance, it looks like the one that appears first in the build file is built first. Before ninja 1.12.0, the one that appears first in the build file is prioritized. This commit floats all affected commands to the top of the ninja build file (as far as it is possible at least) so that older ninja versions will work, and it helps ninja 1.12.0 in the case of ties. A future change will need to be made to artificially lengthen the critical path for affected files so that ninja 1.12.0+ will be fully supported.
jtmorrisbytes
pushed a commit
to jtmorrisbytes/ninja-rs
that referenced
this pull request
May 4, 2026
Add simplified critical path scheduler to improve build times
kinke
added a commit
to kinke/ninja
that referenced
this pull request
Sep 9, 2026
âĶit targets This is a stab at ninja-build#232, allowing to tweak the scheduling order via e.g. `ninja <slow output> all`, for first-time builds without previous build times yet, such as fresh CI builds in ephemeral containers. Implemented by overriding the weight of target-producing edges iff. a) multiple targets have been specified, and b) there is no previous build time for that edge yet. This target-order-specific weight is stored in the upper 32 bits of the 64-bit weight, starting with 1<<32 for the last target's edge, 2<<32 for the 2nd-last target's, 3<<32 for the 3rd-last etc. It should thus always dominate normal weights (e.g., elapsed milliseconds for the previous build). (Something similar to this was implemented in ninja-build#2019, but not in accepted ninja-build#2177. With ninja-build#2682 having landed, I think this is the last remaining interesting piece from ninja-build#2019.)
kinke
added a commit
to kinke/ninja
that referenced
this pull request
Sep 9, 2026
âĶit targets This is a stab at ninja-build#232, allowing to tweak the scheduling order via e.g. `ninja <slow output> all`, for first-time builds without previous build times yet, such as fresh CI builds in ephemeral containers. Implemented by overriding the weight of target-producing edges iff. * multiple targets have been specified, and * there is no previous build time for that edge yet. This target-order-specific weight is stored in the upper 32 bits of the 64-bit weight, starting with 1<<32 for the last target's edge, 2<<32 for the 2nd-last target's, 3<<32 for the 3rd-last etc. It should thus always dominate normal weights (e.g., elapsed milliseconds for the previous build). (Something similar to this was implemented in ninja-build#2019, but not in accepted ninja-build#2177. With ninja-build#2686 having landed, I think this is the last remaining interesting piece from ninja-build#2019.)
kinke
added a commit
to kinke/ninja
that referenced
this pull request
Sep 10, 2026
âĶit targets This is a stab at ninja-build#232, allowing to tweak the scheduling order via e.g. `ninja <slow output> all`, for first-time builds without previous build times yet, such as fresh CI builds in ephemeral containers. Implemented by overriding the weight of target-producing edges iff. * multiple targets have been specified, and * there is no previous build time for that edge yet. This target-order-specific weight is stored in the upper 32 bits of the 64-bit weight, starting with 1<<32 for the last target's edge, 2<<32 for the 2nd-last target's, 3<<32 for the 3rd-last etc. It should thus always dominate normal weights (e.g., elapsed milliseconds for the previous build). (Something similar to this was implemented in ninja-build#2019, but not in accepted ninja-build#2177. With ninja-build#2686 having landed, I think this is the last remaining interesting piece from ninja-build#2019.)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This is a simplified alternative to #2019.
It still adds a critical path scheduler based on weighted edges in the build graph, but instead of using historical runtime it simply assigns a priority of 1 to everything except phony edges. In doing so, it prioritizes jobs by their depth in the build graph. This is better than random scheduling because jobs with dependents will unlock more work to be done when they are completed, thus we get fewer instances of starving the command runner.
For example, if your build has some code-generated sources then the code generator is prioritized above compiling non-generated sources and so all sources files will be unblocked and able to compile in parallel.
Compared to weighting by historic runtime, this will be worse in cases where build times are limited by a single long-running compile job at a low depth and with no dependencies. However, it does avoid the problem of systematically launching the resource-intensive jobs all at once.