irods+phobos = object name must be persistent

[Antoine Migeon]

Hello,

I am in the process of building an iRODS infrastructure composed of "unixfilesystem" resources replicated on "tape" resources for disaster recovery.

I want to reduce costs as much as possible so I don't want to use a proprietary HSM, my first choice was to test dsmarc+univMSS [1].
Today, I plan to try "Phobos" to manage the tapes through a univMSS resource.
Phobos is an opensource project, which is now part of the io-sea project [2].

Like dsmarc, Phobos is not able to rename an archived file: you have to get then put with the new name, which is impossible with many files.
Phobos accepts metadata, but it can't be modified either.
These restrictions exist because most of the information in the database is written to tape in order to be able to retrieve the information on tape in case of database corruption.

My problem is that iRODS does not use a persistent/consistent UUID to name files in the backends (when a user rename/move a file, this file is renamed/moved in backend).
I don't know if there is a way to configure/modify this behavior.

I can choose the name/id of the object when I put it in Phobos, so I'm studying how to get a unique and persistent identifier to use in the univMSS.sh script (syncToArch and stageToCache).

Is the "data_id" database entry persistent for an object?
If yes I suppose I could make a query for each put/get to retrieve this field, and compose a Phobos OID like "zoneName+data_id".
However this solution will not be very efficient by design, is there a better way to do this?


For the moment I'm doing my tests with a "replication" resource, but next I'll try to find a way to do the replication automatically in asynchronous mode, in a delayed way (this will be my first step with iRODS rules!).

Regards,
Antoine

1 : https://github.com/KTH-PDC/irods-dsmarc
2 : https://github.com/phobos-storage

[Kory Draughn]

Hello Antoine,

Yes, the DATA_ID is persistent for data objects. Of course, please test and make sure it enables what you want.

If you come across a case where it doesn't do what you need, please create an issue in the repo. We're happy to think through solutions.

As for whether there is a more efficient design, that will be difficult to answer without measuring performance and considering other factors.

Thanks,

Kory Draughn

Chief Technologist

iRODS Consortium

--
--
The Integrated Rule-Oriented Data System (iRODS) - https://irods.org
 
iROD-Chat: http://groups.google.com/group/iROD-Chat
---
You received this message because you are subscribed to the Google Groups "iRODS-Chat" group.
To unsubscribe from this group and stop receiving emails from it, send an email to irod-chat+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/irod-chat/516e7375-5a6a-4238-85e5-330ad1eddb51n%40googlegroups.com.

[Antoine Migeon]

Hello,

Thank you for your reply. We had put this project on hold, which is why I’m responding so late.

I’m having trouble with “univmss” type composite resources because the filenames are passed as arguments, which causes problems when filenames contains special characters (like single quote).

I saw that it’s possible to modify the physical path (acSetVaultPathPolicy) and add a timestamp to physical object name, which is already helpful and avoids the need for rename/mv operations. But is there a way to change/choose the name of the physical object on vault?

For example, I’d like to encode the file name in Base64 before passing it to univMSS.

Regards,

Antoine

[Bruno Santos]

Hi,

Here is some info on what we have (I might help):

We have this scenario: replication from a unixfilesystem storage to a tape storage.

Handling special chars was one of the issues.

We did it with:
1. acSetVaultPathPolicy set to ramdom scheme.
2. Custom rule to replicate:
2.1. If the file name has special chars not allowed in tape: we rename it to have only allowed chars.
2.2. Replicate to tape.
2.3. If the file was renamed, we restore the original name. This will only update the logical path in the database. The phisical path in the tape resource will contain the chars used in the temporary name.


Example with a file with spaces on the name, where the space is replaced with a _ in tape:

$ ils -L 'br_tst bb'
  user001          1 tape;s3-01           31 2025-04-25.14:04 & br_tst bb
    sha2:ChL/1o9AcGBb8iCa8xJlZjWL4eoTVJwvEcQwDcOCn+Y=    generic    /s3-01/irods/5/12/br_tst_bb.718693.1745582691
$

Regards,

Bruno

[Antoine Migeon]

Hi Bruno,

I hadn't thought of doing it that way—thanks for the tip!
Just to make sure I understand correctly: you rename the "data_name" (logical name) in a rule (or PEP) before running `msisync_to_arch`, and then, once the write operation is complete, you restore the original data_name.
So during step 2.2, the client sees a file with a different name—is that right?

I think I’m still pretty far from understanding how to do that technically... If anyone has an example of a rule for this, that would be great!

Regards,

Antoine

[Kory Draughn]

Antoine,

Perhaps this can aid in what you're attempting to do as well. The project looks to be no more than 2 years old. 

• https://github.com/phobos-storage/irods-phobos

Kory Draughn

Chief Technologist

iRODS Consortium

To view this discussion visit https://groups.google.com/d/msgid/irod-chat/2b185ef9-fb63-4bed-a64e-f6793cc4d500n%40googlegroups.com.

[Antoine Migeon]

Kory,

Yes thank you, this link was in my firt mail.

My problem does not concern Phobos directly. 

It is more a "univmss" problem.

Antoine

You received this message because you are subscribed to a topic in the Google Groups "iRODS-Chat" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/irod-chat/fOUemtxFjCE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to irod-chat+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/irod-chat/CAA-7h7n-oV7RSB30eHmyeo1MBx%3D55jpCvCGF1BgvjbJbeh%3DihA%40mail.gmail.com.

[Kory Draughn]

Ok. Sounds like you already took a look at the univMSS implementation there.

Kory Draughn

Chief Technologist

iRODS Consortium

To view this discussion visit https://groups.google.com/d/msgid/irod-chat/CADU-iVjKwRO49HD3BH%2Bh7_Do9zo8CiNK_UUZVqo-P72EXsXibA%40mail.gmail.com.

[Antoine Migeon]

Yes, I saw their script. In fact, they created it after I told them I wanted to use Phobos behind iRODS a few years ago.

Sorry, I may not have been clear in my explanation. My main concern is the handling of certain characters like single quotes because they can cause problems when passed as arguments to the "univMSS" script. I'm looking for a way to handle this issue before calling the univMSS script.

# upload file on compound resource

$ iput -R comp2 "quote'name.file"
remote addresses: 193.52.249.65 ERROR: putUtil: put error for /mesobfc/home/antoine/quote'name.file, status = -550000 status = -550000 UNIV_MSS_SYNCTOARCH_ERR

$ ils -l "quote'name.file"
  antoine           0 comp2;cache2            7 2026-04-09.22:28 & quote'name.file
  antoine           1 comp2;archive2            7 2026-04-09.22:28 X quote'name.file

# iRODS logs say that the good filename is sent

{"log_category":"legacy","log_level":"info","log_message":"execCmd:/var/lib/irods/msiExecCmd_bin/test_univMSS.sh argv:syncToArch '/irods/storage/1/cache2/home/antoine/quote'name.file' '/irods_phobos/home/antoine/quote'name.file'","request_api_name":"FILE_SYNC_TO_ARCH_AN","request_api_number":525,"request_api_version":"d","request_client_user":"antoine","request_host":"193.52.249.65","request_proxy_user":"rods","request_release_version":"rods5.0.2","server_host":"irods-dev-disk-02.mesobfc.fr","server_pid":3539631,"server_timestamp":"2026-04-09T20:28:02.145Z","server_type":"agent","server_zone":"mesobfc"}

{"log_category":"legacy","log_level":"error","log_message":"[-]\t/irods_source/server/api/src/rsFileSyncToArch.cpp:182:int _rsFileSyncToArch(rsComm_t *, fileStageSyncInp_t *, fileSyncOut_t **) :  status [UNIV_MSS_SYNCTOARCH_ERR]  errno [] -- message [fileSyncToArch failed for [/irods_phobos/home/antoine/quote'name.file]]\n\t[-]\t/irods_source/server/drivers/src/fileDriver.cpp:612:irods::error fileSyncToArch(rsComm_t *, irods::first_class_object_ptr, const std::string &) :  status [UNIV_MSS_SYNCTOARCH_ERR]  errno [] -- message [failed to call 'synctoarch']\n\t\t[-]\t/irods_source/plugins/resources/src/univmss.cpp:648:irods::error univ_mss_file_sync_to_arch(irods::plugin_context &, const char *) :  status [UNIV_MSS_SYNCTOARCH_ERR]  errno [] -- message [univ_mss_file_sync_to_arch: copy of [/irods/storage/1/cache2/home/antoine/quote'name.file] to [/irods_phobos/home/antoine/quote'name.file] failed.   stdout buff [0]   stderr buff [0x55d2ca8315b0]  status [-344000]]\n\n","request_api_name":"FILE_SYNC_TO_ARCH_AN","request_api_number":525,"request_api_version":"d","request_client_user":"antoine","request_host":"193.52.249.65","request_proxy_user":"rods","request_release_version":"rods5.0.2","server_host":"irods-dev-disk-02.mesobfc.fr","server_pid":3539626,"server_timestamp":"2026-04-09T20:28:07.152Z","server_type":"agent","server_zone":"mesobfc"}

# but univMSS script receive wrong args :

$1 == syncToArch

$2 == /irods/storage/1/cache2/home/antoine/quote

$3 == name.file

Sincerely,

[Kory Draughn]

Antoine,

I've reproduced the problem and I think it's a bug. I'm going to do more testing so that I'm sure.

If it turns out to be a real bug, I'll open an issue for it.

Kory Draughn

Chief Technologist

iRODS Consortium

To view this discussion visit https://groups.google.com/d/msgid/irod-chat/4f10e08c-0f12-4951-9d44-942af38bf683n%40googlegroups.com.

[Kory Draughn]

Antoine,

I opened https://github.com/irods/irods/issues/8928 to address the bug. Feel free to leave any additional comments in the issue.

The fix will be included in the next release of the server.

Thanks,

Kory Draughn

Chief Technologist

iRODS Consortium

[Bruno Freire Santos]

Hi Antoine,

Sorry for the late reply.

Just to complete the answer:

Yes, the rule works as you describe.

It is a rule triggered from a cron and not via peps, it selects data objects to replicate based on metadata and triggers the replication.

We had a recent issue and I made a small description of the full rule here:

https://groups.google.com/g/iROD-Chat/c/O-t7XffGsAg

The rename is not described there, as it a special condition on the replicate.

And yes, during the replication the user will see the temporary name. Normally it is not that long.

Regards

To view this discussion visit https://groups.google.com/d/msgid/irod-chat/CAA-7h7nj3qojCd%3Drdob9wy%3DsovJw%3DoxETJsXHocTQZ_4a0xyFQ%40mail.gmail.com.

[Antoine Migeon]

Hello Bruno and Kory,

Thank you so much for your help, and thank you for working while I was on vacation!

@Kory, this fix will certainly resolve a large part of my issue. And I see that the fix has already been made on GitHub—well done! Is there any chance it will be included in a 5.0.X release?

Regardless of this univMSS issue (I'm asking one last time, sorry for being so insistent): is there a way to choose the name of the file written to the resource (for example in unixfilesystem mode), perhaps at the beginning of the put operation with a pre PEP or when creating the entry to the database?

@Bruno, I'm not sure this solution works for all our use cases (partly because writing to tape can be time-consuming and delayed). Would you be able to share the entire rule (perhaps privately)?

Regards,

Antoine

[Kory Draughn]

Antoine,

Sorry for the late response.

@Kory, this fix will certainly resolve a large part of my issue. And I see that the fix has already been made on GitHub—well done! Is there any chance it will be included in a 5.0.X release?

There are no plans to release a 5.0.3. The upcoming 5.1.0 release fixes bugs and adds functionality. Upgrading to it should be smooth.

Regardless of this univMSS issue (I'm asking one last time, sorry for being so insistent): is there a way to choose the name of the file written to the resource (for example in unixfilesystem mode), perhaps at the beginning of the put operation with a pre PEP or when creating the entry to the database?

I'm pretty sure what you're asking for can be done via a custom C++ plugin. That is because plugins have direct access to the API inputs. Of course, that would require someone to write C++.

If we were to offer a way to adjust the filename at the physical level, we'd likely do that through a resource plugin because it allows us to keep the logic in one place.

The recent fix for the univmss resource plugin coupled with msiSetRandomScheme() could be enough to satisfy your needs. That unfortunately means waiting until 5.1.0 is released.

There might be another way to achieve what you want through the use of a new microservice. Again, this would be a 5.1.0 thing. Consider the following from core.re.template.

    # acSetVaultPathPolicy {msiSetRandomScheme; }

    acSetVaultPathPolicy {msiSetGraftPathScheme("no","1"); }

acSetVaultPathPolicy is the hook which controls the path scheme. We should be able to implement a new microservice which instructs iRODS to call a rule/function which you control/implement. That would give you the ability to modify the path or filename. If that sounds useful, we can look into getting that in for the release.

Questions for you:

• What organization are you with?
• You mentioned encoding the filename using base64. Does that mean reversing the encoding is desired?
• Do you need to modify the full path or just the filename?
• Do you need full control over the physical path? Can the problem be solved without full control over the path?
• Is there a fixed set of options which can achieve what you want?
• e.g. Appending the data id, replacing a set of characters with one character, using UUIDs
• Are there other features you feel would be helpful in resolving this?

Overall, I don't think we can achieve the solution you want without releasing a new version of iRODS. And outside of the solution from Bruno, I think your only option (until 5.1.0) is to reject paths which contain problematic characters. 

Kory Draughn

Chief Technologist

iRODS Consortium

To view this discussion visit https://groups.google.com/d/msgid/irod-chat/bb6add8e-5947-4b74-b32d-e8a95469b258n%40googlegroups.com.

[Antoine Migeon]

Hello Kory,

Thank you for your responses.
My turn to apologize for the long delay in responding.

I'm pretty sure what you're asking for can be done via a custom C++ plugin. That is because plugins have direct access to the API inputs. Of course, that would require someone to write C++.

If we were to offer a way to adjust the filename at the physical level, we'd likely do that through a resource plugin because it allows us to keep the logic in one place.

I’d prefer to avoid a locally developed solution that wouldn’t be maintained by the community, since we unfortunately don’t have enough staff to sustain that kind of initiative.
 

 

The recent fix for the univmss resource plugin coupled with msiSetRandomScheme() could be enough to satisfy your needs. That unfortunately means waiting until 5.1.0 is released.

Yes, it might be possible for us to start that way; I still need to analyze the data from the existing system in greater detail and run more tests to see if we’ll run into any issues.

acSetVaultPathPolicy is the hook which controls the path scheme. We should be able to implement a new microservice which instructs iRODS to call a rule/function which you control/implement. That would give you the ability to modify the path or filename. If that sounds useful, we can look into getting that in for the release.

Yes I took a look at acSetVaultPathPolicy and use it with msiSetRandomScheme, which had already provided part of the solution regarding the file path and also helps avoid renaming operations.
A new microservice that would allow us to modify the file name in addition to the path would be perfect.

 

Questions for you:

• What organization are you with?

We are the computing center at the "Université Bourgogne Europe" in Dijon, France; our mission is to provide computing and storage resources for the university's research laboratories.. https://www.ube.fr 

• You mentioned encoding the filename using base64. Does that mean reversing the encoding is desired?

 I think being able to decode the name could be useful in some cases, such as when the database is corrupted, but it's not required.

• Do you need to modify the full path or just the filename?

In my case, the full path should be processed, but most of the work is already done by msiSetRandomScheme. 

• Do you need full control over the physical path? Can the problem be solved without full control over the path? 

• • Is there a fixed set of options which can achieve what you want?
    • e.g. Appending the data id, replacing a set of characters with one character, using UUIDs

No, I don't need to have complete control.

 In my opinion, the simplest and most effective solution would be:
- path using the msiSetRandomScheme form = /vault/.../random_integer/random-integer/
- filename using the form = data_id.timestamp

If I understand correctly, the data_id is persistent and consistent for each file and will never change, so it's already a kind of UUID. So I think using the ID instead of the name would be enough (not appending).

• Are there other features you feel would be helpful in resolving this?

Overall, I don't think we can achieve the solution you want without releasing a new version of iRODS. And outside of the solution from Bruno, I think your only option (until 5.1.0) is to reject paths which contain problematic characters. 

I don't think I need any other features.

Yes, I understand. I’ll follow your advice and discard everything I can, but I need to migrate from an existing system (HPE/SGI DMF) with tens of millions of files, so it’s no easy task!

Thank you for your help Kory.

And thank you as well, Bruno, for the help you offered me. I didn’t pursue that route because I don’t think that solution is ideal, so I’ll consider it as a last resort.

Regards,
Antoine