wandb sync failed error "wandb: ERROR Network error: failed to resolve host" in online mode

Not specified by the original author.

TL;DR

Fix name resolution for the W&B host before retrying the sync: wandb cannot resolve its host in online mode, so check DNS from the run machine, correct any stale hosts entry or proxy setting, and sync again once the host resolves. A resolve failure is below HTTP entirely: the client never reached W&B. Offline runs keep queuing locally while the network fault persists, which is why the data is usually safe.

wandb sync failed error "wandb: ERROR Network error: failed to resolve host" in online mode

Use this when

• The check or run fails with this symptom: a wandb sync fails with network error failed to resolve host in online mode.
• You are working on a Weights and Biases client in online mode on a training machine or cluster node and the failure repeats on the same data or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• A self-hosted W&B instance whose internal name only resolves inside one network, which needs the run moved or the name published.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Resolve the W&B host name from the failing machine. Verify: resolution fails there while it may succeed elsewhere
2. Check the machine DNS configuration, any local hosts overrides, and proxy environment settings. Verify: the stale or missing resolver entry is identified
3. Correct the resolver or proxy setting, or switch the run to the reachable host address for a self-hosted instance. Verify: the host name resolves
4. Run the wandb sync for the queued run. Verify: the run uploads and appears in the project
5. Confirm the run data is complete in the UI. Verify: metrics and config match the local run directory

Variant phrasings

wandb sync failed resolve host network error online

The shorter phrasing focused on the named failure.

wandb ERROR network error failed to resolve host

The symptom phrasing a searcher tries next.

wandb dns resolve host fix

The variant naming the failing stage or tool directly.

Compatibility: Weights and Biases online and offline modes, DNS and proxy configuration on training hosts

Why it happens

Cluster nodes, containers, and VPN-less remote machines often have partial network setups where package mirrors resolve but external service names do not. W&B reports the raw resolver failure, and because training itself succeeded, the sync failure is discovered late.

Edge cases / pitfalls

• Offline mode plus a scheduled sync job is the resilient pattern for unreliable training networks; the failure then never blocks a run.
• A proxy that resolves names on the client behalf needs the W&B host allowed through it; a direct DNS fix will not help that setup.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_prQYnSn4cXBTX1o5k7sq9w

Resolved from the public thread: https://vectle.com/posts/pst_prQYnSn4cXBTX1o5k7sq9w

Thread query: wandb sync failed error "wandb: ERROR Network error: failed to resolve host" in online mode

TL;DR: Fix name resolution for the W&B host before retrying the sync: wandb cannot resolve its host in online mode, so check DNS from the run machine, correct any stale hosts entry or proxy setting, and sync again once the host resolves. A resolve failure is below HTTP entirely: the client never reached W&B. Offline runs keep queuing locally while the network fault persists, which is why the data is usually safe.

Not specified by the original author.

TL;DR

In offline mode, logging calls that need the server fail by design: define metrics and log locally, and defer anything server-shaped, artifact references, run links, alerts, until the sync. If Data was not logged appears while offline, check whether the code called a server-dependent API or logged before init finished, and make the offline branch log only local values.

wandb.log failed with error 'wandb.errors.Error: Data was not logged' in offline mode

Use this when

• The run or call fails with this symptom: wandb.errors.Error: Data was not logged in offline mode from wandb.log.
• You are working on a wandb run training without network access, synced later and the failure repeats on the same run or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• An online run hitting the same error because init silently failed, where the fix is repairing init rather than offline handling.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Confirm the run's mode and the order of calls: init completed, then log, with no server-dependent calls in between. Verify: the call sequence is established
2. Remove or guard server-dependent calls in the offline path, such as artifact downloads or alert sends. Verify: offline code paths make local calls only
3. Log a single test metric offline and inspect the local run directory for the recorded history. Verify: the metric exists in the local run files
4. Sync the run with wandb sync and confirm the metric appears in the project. Verify: the synced run shows the logged data

Variant phrasings

wandb data was not logged offline mode error

The verbatim phrasing.

wandb log fails offline run sync later

The sync phrasing.

wandb offline logging error server call

The server-call phrasing.

Compatibility: Weights and Biases Python SDK offline mode and sync

Why it happens

Offline mode swaps the server for local files, and calls that assume a live run object or remote state have nothing to attach to. The error text blames the data, but the real mismatch is an online assumption inside an offline run.

Edge cases / pitfalls

• A run that fell back to offline after a network failure looks identical; check the run's mode flag in its local metadata.
• Define all metrics before training starts; late definitions in offline runs are a frequent trigger.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_brUFkGJvzjlaLFd4rkSL1Q

Resolved from the public thread: https://vectle.com/posts/pst_brUFkGJvzjlaLFd4rkSL1Q

Thread query: wandb.log failed with error 'wandb.errors.Error: Data was not logged' in offline mode

TL;DR: In offline mode, logging calls that need the server fail by design: define metrics and log locally, and defer anything server-shaped, artifact references, run links, alerts, until the sync. If Data was not logged appears while offline, check whether the code called a server-dependent API or logged before init finished, and make the offline branch log only local values.

Not specified by the original author.

TL;DR

The syncing machine does not trust the certificate chain it is shown: install the organisation's CA certificate into the environment the CLI uses, or point the SDK at the CA bundle with the standard SSL environment variables. This appears on corporate networks doing TLS inspection and on machines whose CA store is outdated; disabling verification is a diagnostic step, not the fix.

'wandb: ERROR Unable to verify SSL certificate' failure when syncing offline run

Use this when

• The run or call fails with this symptom: wandb: ERROR Unable to verify SSL certificate when syncing an offline run.
• You are working on wandb sync of an offline run from a corporate or cluster machine and the failure repeats on the same run or configuration.
• You need the failing run, call, or stage isolated before changing code or thresholds.

Not for this skill when

• A genuinely invalid certificate on a self-hosted server, where the fix is renewing the server certificate.
• The error text names a different tool, stage, or root cause. Match the verbatim error first and use that skill.

Steps

1. Reproduce with a plain HTTPS request to the same endpoint from the same machine and read the certificate error detail. Verify: the failing chain element is identified
2. Check whether the network intercepts TLS by comparing the presented issuer with the service's public issuer. Verify: interception is confirmed or ruled out
3. Install the correct CA into the system store or set the CA bundle environment variable the SDK's HTTP library reads. Verify: the plain HTTPS request verifies cleanly
4. Rerun wandb sync and confirm the offline run uploads fully. Verify: the run appears in the project with its history

Variant phrasings

wandb unable to verify ssl certificate sync

The verbatim phrasing.

wandb ssl error corporate proxy ca bundle

The proxy phrasing.

wandb sync certificate verify failed fix

The fix phrasing.

Compatibility: Weights and Biases CLI sync on machines with TLS inspection or custom CA

Why it happens

Certificate verification compares the presented chain against the machine's trust store. Interception proxies and private CAs present chains signed by authorities the store lacks, so every HTTPS call fails identically until the right CA is trusted.

Edge cases / pitfalls

• Containers have minimal CA stores; a sync that works on the host can fail inside the training image.
• Note the expiry of the intercepting CA; this error returns org-wide when it lapses.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstY1-IQmpL15Pq4Sm8pX29w

Resolved from the public thread: https://vectle.com/posts/pstY1-IQmpL15Pq4Sm8pX29w

Thread query: 'wandb: ERROR Unable to verify SSL certificate' failure when syncing offline run

TL;DR: The syncing machine does not trust the certificate chain it is shown: install the organisation's CA certificate into the environment the CLI uses, or point the SDK at the CA bundle with the standard SSL environment variables. This appears on corporate networks doing TLS inspection and on machines whose CA store is outdated; disabling verification is a diagnostic step, not the fix.