RFR: 8367005: ImageReader refactor caused performance regressions for startup and footprint [v2]

[Claes Redestad]

On Wed, 10 Sep 2025 22:05:26 GMT, David Beaumont <duke at openjdk.org> wrote:

>> Summary: Add two new methods to ImageReader to make SystemModuleReader more performant.
>> 
>> Analysis of benchmarks shows that when the vast majority of resources (~11,000 in the Perfstartup-SwingSet-G1 benchmark) tested for in SystemModuleReader do NOT exist, performance is degraded compared to this code prior to the refactoring in JDK-8360037.
>> 
>> The current refactoring of ImageReader has everything going through a single "findNode()" method for simplest possible encapsulation, but while this is functionally correct, it's not tuned for testing for the non-existence of resources.
>> 
>> In particular:
>> 1. SystemModuleReader only requests resources (i.e. things in the jimage file with paths *not* starting /modules/ or /packages/). This means findNode() does two look-ups for the resource, the first of which will always fail.
>> 2. The containsResource() logic doesn't need to create and cache nodes in ImageReader, it can just check for the presence of an ImageLocation corresponding to a resource.
>> 
>> Thus two new methods are added to resolve these cases:
>> * findResourceNode(module, path)
>> * containsResource(module, path)
>> 
>> Their API takes module and path separately so as to not be confusable with findNode(nodename).
>> 
>> However care must be taken to prevent these methods being fooled into returning non-resource entries (this was possible before the refactoring by using module names like "modules" or "packages") so new tests have been added.
>
> David Beaumont has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Found additional place where new API can be used.

src/java.base/share/classes/jdk/internal/jimage/ImageReader.java line 379:

> 377:             // exist, so it skips checking the nodes cache and only checks for
> 378:             // an ImageLocation.
> 379:             if (moduleName.contains("/") || resourcePath.startsWith("/")) {

The fast-path guard here and above might improve if they were done before synchronizing, e.g. dropping the modifier and putting a `synchonized (this)` block around the rest of the method. 

Putting the likely cheaper `resourcePath.startsWith` first might be a small win, too, depending on input/benchmark.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/27203#discussion_r2338193115