update_physical_path_for_decoupled_naming() is unnecessarily executed twice when there is no redirect

[JustinKyleJames]

• main

---

In the redirect scenario, update_physical_path_for_decoupled_naming must be called from two different locations.

1. From s3_notify which is run on the server the client is connected to. This ensures that the decoupled name get written properly to the database.
2. From s3_file_create which is run on the server we are redirected to. This ensures that object is written to the correct S3 key.

In the non-redirect case, the above two code paths still both call update_physical_path_for_decoupled_naming. This means that the code will iterate through the L1desc[] table twice (usually only one or two iterations as the entries are almost always in the first couple of slots) and update the physical_path twice. This is slightly inefficient.

We should detect that this has already been executed on that server and not execute it a second time.

[trel]

nice.

so we save a value somewhere in step 1, and when in step 2, we should detect/check that value and skip going to the database.

[JustinKyleJames]

nice.

so we save a value somewhere in step 1, and when in step 2, we should detect/check that value and skip going to the database.

Well, in the non-redirect case we will never go to the database because the L1desc[] table is available.

[trel]

right.