Repository navigation
Description
OpenNebula oVirt API wrapper (ovirtapi-server, opennebula-ovirtapi-7.4.0-1.el9 + the 07.09.26 patch), compared against OLVM.
Summary
Requests for unknown paths, unimplemented sub-collections and unknown VM or cluster ids return HTTP 500 Internal Server Error instead of 404 Not Found. The error bodies use the correct oVirt fault format; only the status code is wrong. The underlying Sinatra::NotFound and "not found" lookups are reported as server errors.
Behaviour is also inconsistent across collections, and a malformed id silently resolves to host 0.
Note that GET /vms/00000000-0000-0000-0000-000000000000 correctly returns 404 VM 0 not found, while a different unknown VM id returns 500.
Response examples
OpenNebula GET /ovirt-engine/api/nonexistent:
HTTP/1.1 500 Internal Server Error Content-Type: application/json
{"fault":{"reason":"Operation Failed","detail":"Internal server error. (Sinatra::NotFound: Sinatra::NotFound)"}}
OpenNebula GET /ovirt-engine/api/vms/99999999-0000-0000-0000-000000000000:
HTTP/1.1 500 Internal Server Error
{"fault":{"reason":"Operation Failed","detail":"Unable to retrieve VM 99999999000000000000000000000000"}}
OpenNebula GET /ovirt-engine/api/hosts/not-a-uuid:
HTTP/1.1 200 OK
{"name":"opennebula-node1.robofish.local","address":"opennebula-node1.robofish.local","id":"00000000-0000-0000-0000-000000000000", …}
OLVM, same kind of requests:
GET /ovirt-engine/api/nonexistent
HTTP/2 404
Content-Type: text/html;charset=UTF-8RESTEASY003210: Could not find resource for full path: https://…/ovirt-engine/api/v4/nonexistent
GET /ovirt-engine/api/hosts/00000000-0000-0000-0000-000000000000
HTTP/2 404
Content-Length: 0
Impact
A client can't tell "object doesn't exist" from "server failed". A 500 is typically retried, so a deleted VM or cluster turns into a retry loop and a misleading "server error" in logs instead of a clean "not found". The not-a-uuid → host 0 case can make a malformed id act on the wrong host.
Expected
404 (oVirt
faultbody) for unknown paths, unimplemented sub-collections and unknown or malformed ids; 5xx only for real server failures.Don't advertise links (
host.link[]) to sub-collections that aren't implemented.
Logs
onl-status-codes-evidence.zip contains:
onl-status-requests.txt: raw requests and responses (headers + body) for every row above. Session cookies are redacted.olvm-reference-requests.txt: the OLVM equivalents for comparison.
Found while investigating #32.
Tracked as TFS #1731095 (Risk — Severity 4 - Low).
OpenNebula oVirt API wrapper (ovirtapi-server, opennebula-ovirtapi-7.4.0-1.el9 + the 07.09.26 patch), compared against OLVM.
Summary
Requests for unknown paths, unimplemented sub-collections and unknown VM or cluster ids return HTTP 500 Internal Server Error instead of 404 Not Found. The error bodies use the correct oVirt fault format; only the status code is wrong. The underlying Sinatra::NotFound and "not found" lookups are reported as server errors.
Behaviour is also inconsistent across collections, and a malformed id silently resolves to host 0.
| Request | OpenNebula | OLVM (expected) |
|---|---|---|
GET /nonexistent |
500 Sinatra::NotFound |
404 |
GET /hosts/{id}/statistics (advertised in host.link[]) |
500 Sinatra::NotFound |
200 with data (see #32) |
GET /hosts/{id}/nics (advertised in host.link[]) |
500 Sinatra::NotFound |
200 with data |
GET /vms/99999999-0000-0000-0000-000000000000 |
500 Unable to retrieve VM 9999… |
404 |
GET /clusters/99999999-0000-0000-0000-000000000000 |
500 Error obtaining OpenNebula cluster: 9999… |
404 |
GET /hosts/99999999-0000-0000-0000-000000000000 |
404 {} ✅ (but an empty body, no fault) |
404 |
GET /hosts/not-a-uuid |
200, returns host 0 (opennebula-node1) |
404 |
Note that GET /vms/00000000-0000-0000-0000-000000000000 correctly returns 404 VM 0 not found, while a different unknown VM id returns 500.
Response examples
OpenNebula GET /ovirt-engine/api/nonexistent:
HTTP/1.1 500 Internal Server Error
Content-Type: application/json
{"fault":{"reason":"Operation Failed","detail":"Internal server error. (Sinatra::NotFound: Sinatra::NotFound)"}}
OpenNebula GET /ovirt-engine/api/vms/99999999-0000-0000-0000-000000000000:
HTTP/1.1 500 Internal Server Error
{"fault":{"reason":"Operation Failed","detail":"Unable to retrieve VM 99999999000000000000000000000000"}}
OpenNebula GET /ovirt-engine/api/hosts/not-a-uuid:
HTTP/1.1 200 OK
{"name":"opennebula-node1.robofish.local","address":"opennebula-node1.robofish.local","id":"00000000-0000-0000-0000-000000000000", …}
OLVM, same kind of requests:
GET /ovirt-engine/api/nonexistent
HTTP/2 404
Content-Type: text/html;charset=UTF-8
RESTEASY003210: Could not find resource for full path: https://…/ovirt-engine/api/v4/nonexistent
GET /ovirt-engine/api/hosts/00000000-0000-0000-0000-000000000000
HTTP/2 404
Content-Length: 0
Impact
A client can't tell "object doesn't exist" from "server failed". A 500 is typically retried, so a deleted VM or cluster turns into a retry loop and a misleading "server error" in logs instead of a clean "not found". The not-a-uuid → host 0 case can make a malformed id act on the wrong host.
Expected
- 404 (oVirt
faultbody) for unknown paths, unimplemented sub-collections and unknown or malformed ids; 5xx only for real server failures. - Don't advertise links (
host.link[]) to sub-collections that aren't implemented.
Activity